d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
69cdfcef2e
freebsd-skq/sys/dev/txp/if_txp.c

3019 lines
86 KiB
C
Raw Normal View History

You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
/* $OpenBSD: if_txp.c,v 1.48 2001/06/27 06:34:50 kjc Exp $ */
Start each of the license/copyright comments with /*-, minor shuffle of lines
2005-01-06 01:43:34 +00:00
/*-
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
* Copyright (c) 2001
* Jason L. Wright <jason@thought.net>, Theo de Raadt, and
* Aaron Campbell <aaron@monkey.org>. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by Jason L. Wright,
* Theo de Raadt and Aaron Campbell.
* 4. Neither the name of the author nor the names of any co-contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
* THE POSSIBILITY OF SUCH DAMAGE.
*/
Use __FBSDID(). Also some minor style cleanups.
2003-08-24 17:55:58 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
/*
* Driver for 3c990 (Typhoon) Ethernet ASIC
*/
#include <sys/param.h>
#include <sys/systm.h>
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#include <sys/bus.h>
#include <sys/endian.h>
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
#include <sys/kernel.h>
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
Add missing <sys/module.h> includes
2004-05-30 20:08:47 +00:00
#include <sys/module.h>
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#include <sys/mutex.h>
#include <sys/queue.h>
#include <sys/rman.h>
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
#include <sys/socket.h>
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#include <sys/sockio.h>
#include <sys/sysctl.h>
#include <sys/taskqueue.h>
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#include <net/bpf.h>
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
#include <net/if.h>
The r48589 promised to remove implicit inclusion of if_var.h soon. Prepare to this event, adding if_var.h to files that do need it. Also, include all includes that now are included due to implicit pollution via if_var.h Sponsored by: Netflix Sponsored by: Nginx, Inc.
2013-10-26 17:58:36 +00:00
#include <net/if_var.h>
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
#include <net/if_arp.h>
#include <net/ethernet.h>
#include <net/if_dl.h>
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#include <net/if_media.h>
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
#include <net/if_types.h>
Make vlan(4) loadable, unloadable, and clonable. As a side effect, interfaces must now always enable VLAN support. Reviewed by: jlemon MFC after: 3 weeks
2001-09-05 21:10:28 +00:00
#include <net/if_vlan_var.h>
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/ip.h>
#include <dev/mii/mii.h>
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
#include <dev/pci/pcireg.h>
#include <dev/pci/pcivar.h>
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#include <machine/bus.h>
#include <machine/in_cksum.h>
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
#include <dev/txp/if_txpreg.h>
#include <dev/txp/3c990img.h>
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
MODULE_DEPEND(txp, pci, 1, 1, 1);
MODULE_DEPEND(txp, ether, 1, 1, 1);
/*
* XXX Known Typhoon firmware issues.
*
* 1. It seems that firmware has Tx TCP/UDP checksum offloading bug.
* The firmware hangs when it's told to compute TCP/UDP checksum.
* I'm not sure whether the firmware requires special alignment to
* do checksum offloading but datasheet says nothing about that.
* 2. Datasheet says nothing for maximum number of fragmented
* descriptors supported. Experimentation shows up to 16 fragment
* descriptors are supported in the firmware. For TSO case, upper
* stack can send 64KB sized IP datagram plus link header size(
* ethernet header + VLAN tag) frame but controller can handle up
* to 64KB frame given that PAGE_SIZE is 4KB(i.e. 16 * PAGE_SIZE).
* Because frames that need TSO operation of hardware can be
* larger than 64KB I disabled TSO capability. TSO operation for
* less than or equal to 16 fragment descriptors works without
* problems, though.
* 3. VLAN hardware tag stripping is always enabled in the firmware
* even if it's explicitly told to not strip the tag. It's
* possible to add the tag back in Rx handler if VLAN hardware
* tag is not active but I didn't try that as it would be
* layering violation.
* 4. TXP_CMD_RECV_BUFFER_CONTROL does not work as expected in
* datasheet such that driver should handle the alignment
* restriction by copying received frame to align the frame on
* 32bit boundary on strict-alignment architectures. This adds a
* lot of CPU burden and it effectively reduce Rx performance on
Remove ia64. This includes: o All directories named *ia64* o All files named *ia64* o All ia64-specific code guarded by __ia64__ o All ia64-specific makefile logic o Mention of ia64 in comments and documentation This excludes: o Everything under contrib/ o Everything under crypto/ o sys/xen/interface o sys/sys/elf_common.h Discussed at: BSDcan
2014-07-07 00:27:09 +00:00
* strict-alignment architectures(e.g. sparc64, arm and mips).
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
*
* Unfortunately it seems that 3Com have no longer interests in
* releasing fixed firmware so we may have to live with these bugs.
*/
#define TXP_CSUM_FEATURES (CSUM_IP)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
/*
* Various supported device vendors/types and their names.
*/
static struct txp_type txp_devs[] = {
{ TXP_VENDORID_3COM, TXP_DEVICEID_3CR990_TX_95,
"3Com 3cR990-TX-95 Etherlink with 3XP Processor" },
{ TXP_VENDORID_3COM, TXP_DEVICEID_3CR990_TX_97,
"3Com 3cR990-TX-97 Etherlink with 3XP Processor" },
{ TXP_VENDORID_3COM, TXP_DEVICEID_3CR990B_TXM,
"3Com 3cR990B-TXM Etherlink with 3XP Processor" },
{ TXP_VENDORID_3COM, TXP_DEVICEID_3CR990_SRV_95,
"3Com 3cR990-SRV-95 Etherlink Server with 3XP Processor" },
{ TXP_VENDORID_3COM, TXP_DEVICEID_3CR990_SRV_97,
"3Com 3cR990-SRV-97 Etherlink Server with 3XP Processor" },
{ TXP_VENDORID_3COM, TXP_DEVICEID_3CR990B_SRV,
"3Com 3cR990B-SRV Etherlink Server with 3XP Processor" },
{ 0, 0, NULL }
};
- Only call txp_release_resources() once if attach fails. - Set errno to ENXIO instead of 0 in several attach failure cases. - Setup the interrupt handler at the very end of txp_attach() after ether_ifattach(). - Various whitespace fixes in function prototypes.
2005-08-31 18:09:54 +00:00
static int txp_probe(device_t);
static int txp_attach(device_t);
static int txp_detach(device_t);
static int txp_shutdown(device_t);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static int txp_suspend(device_t);
static int txp_resume(device_t);
static int txp_intr(void *);
static void txp_int_task(void *, int);
static void txp_tick(void *);
- Only call txp_release_resources() once if attach fails. - Set errno to ENXIO instead of 0 in several attach failure cases. - Setup the interrupt handler at the very end of txp_attach() after ether_ifattach(). - Various whitespace fixes in function prototypes.
2005-08-31 18:09:54 +00:00
static int txp_ioctl(struct ifnet *, u_long, caddr_t);
- Provide txp_get_counter() to return counters that are not collected, but taken from hardware. - Mechanically convert to if_inc_counter() the rest of counters.
2014-09-24 11:58:23 +00:00
static uint64_t txp_get_counter(struct ifnet *, ift_counter);
- Only call txp_release_resources() once if attach fails. - Set errno to ENXIO instead of 0 in several attach failure cases. - Setup the interrupt handler at the very end of txp_attach() after ether_ifattach(). - Various whitespace fixes in function prototypes.
2005-08-31 18:09:54 +00:00
static void txp_start(struct ifnet *);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
static void txp_start_locked(struct ifnet *);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static int txp_encap(struct txp_softc *, struct txp_tx_ring *, struct mbuf **);
- Only call txp_release_resources() once if attach fails. - Set errno to ENXIO instead of 0 in several attach failure cases. - Setup the interrupt handler at the very end of txp_attach() after ether_ifattach(). - Various whitespace fixes in function prototypes.
2005-08-31 18:09:54 +00:00
static void txp_stop(struct txp_softc *);
static void txp_init(void *);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
static void txp_init_locked(struct txp_softc *);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static void txp_watchdog(struct txp_softc *);
Remove __P.
2002-03-20 02:08:01 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static int txp_reset(struct txp_softc *);
static int txp_boot(struct txp_softc *, uint32_t);
static int txp_sleep(struct txp_softc *, int);
static int txp_wait(struct txp_softc *, uint32_t);
Remove __P.
2002-03-20 02:08:01 +00:00
static int txp_download_fw(struct txp_softc *);
static int txp_download_fw_wait(struct txp_softc *);
- Only call txp_release_resources() once if attach fails. - Set errno to ENXIO instead of 0 in several attach failure cases. - Setup the interrupt handler at the very end of txp_attach() after ether_ifattach(). - Various whitespace fixes in function prototypes.
2005-08-31 18:09:54 +00:00
static int txp_download_fw_section(struct txp_softc *,
Remove __P.
2002-03-20 02:08:01 +00:00
struct txp_fw_section_header *, int);
static int txp_alloc_rings(struct txp_softc *);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static void txp_init_rings(struct txp_softc *);
static int txp_dma_alloc(struct txp_softc *, char *, bus_dma_tag_t *,
bus_size_t, bus_size_t, bus_dmamap_t *, void **, bus_size_t, bus_addr_t *);
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
static void txp_dma_free(struct txp_softc *, bus_dma_tag_t *, bus_dmamap_t,
void **, bus_addr_t *);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static void txp_free_rings(struct txp_softc *);
Remove __P.
2002-03-20 02:08:01 +00:00
static int txp_rxring_fill(struct txp_softc *);
static void txp_rxring_empty(struct txp_softc *);
static void txp_set_filter(struct txp_softc *);
static int txp_cmd_desc_numfree(struct txp_softc *);
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
static int txp_command(struct txp_softc *, uint16_t, uint16_t, uint32_t,
uint32_t, uint16_t *, uint32_t *, uint32_t *, int);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static int txp_ext_command(struct txp_softc *, uint16_t, uint16_t,
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint32_t, uint32_t, struct txp_ext_desc *, uint8_t,
Remove __P.
2002-03-20 02:08:01 +00:00
struct txp_rsp_desc **, int);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static int txp_response(struct txp_softc *, uint16_t, uint16_t,
Remove __P.
2002-03-20 02:08:01 +00:00
struct txp_rsp_desc **);
- Only call txp_release_resources() once if attach fails. - Set errno to ENXIO instead of 0 in several attach failure cases. - Setup the interrupt handler at the very end of txp_attach() after ether_ifattach(). - Various whitespace fixes in function prototypes.
2005-08-31 18:09:54 +00:00
static void txp_rsp_fixup(struct txp_softc *, struct txp_rsp_desc *,
Remove __P.
2002-03-20 02:08:01 +00:00
struct txp_rsp_desc *);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static int txp_set_capabilities(struct txp_softc *);
Remove __P.
2002-03-20 02:08:01 +00:00
static void txp_ifmedia_sts(struct ifnet *, struct ifmediareq *);
static int txp_ifmedia_upd(struct ifnet *);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
#ifdef TXP_DEBUG
Remove __P.
2002-03-20 02:08:01 +00:00
static void txp_show_descriptor(void *);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
#endif
Remove __P.
2002-03-20 02:08:01 +00:00
static void txp_tx_reclaim(struct txp_softc *, struct txp_tx_ring *);
static void txp_rxbuf_reclaim(struct txp_softc *);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#ifndef __NO_STRICT_ALIGNMENT
static __inline void txp_fixup_rx(struct mbuf *);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
#endif
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static int txp_rx_reclaim(struct txp_softc *, struct txp_rx_ring *, int);
static void txp_stats_save(struct txp_softc *);
static void txp_stats_update(struct txp_softc *, struct txp_rsp_desc *);
static void txp_sysctl_node(struct txp_softc *);
static int sysctl_int_range(SYSCTL_HANDLER_ARGS, int, int);
static int sysctl_hw_txp_proc_limit(SYSCTL_HANDLER_ARGS);
static int prefer_iomap = 0;
TUNABLE_INT("hw.txp.prefer_iomap", &prefer_iomap);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
static device_method_t txp_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, txp_probe),
DEVMETHOD(device_attach, txp_attach),
DEVMETHOD(device_detach, txp_detach),
DEVMETHOD(device_shutdown, txp_shutdown),
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
DEVMETHOD(device_suspend, txp_suspend),
DEVMETHOD(device_resume, txp_resume),
{ NULL, NULL }
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
};
static driver_t txp_driver = {
"txp",
txp_methods,
sizeof(struct txp_softc)
};
static devclass_t txp_devclass;
- Express hard dependencies on bus (pci, isa, pccard) and network layer (ether). - Don't abuse module names to facilitate ifconfig module loading; such abuse isn't really needed. (And if we do need type information associated with a module then we should make it explicit and not use hacks.)
2003-04-15 06:37:30 +00:00
DRIVER_MODULE(txp, pci, txp_driver, txp_devclass, 0, 0);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_probe(device_t dev)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
struct txp_type *t;
t = txp_devs;
style(9) - space after keywords.
2009-03-11 08:49:17 +00:00
while (t->txp_name != NULL) {
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if ((pci_get_vendor(dev) == t->txp_vid) &&
(pci_get_device(dev) == t->txp_did)) {
device_set_desc(dev, t->txp_name);
style(9) - space after keywords.
2009-03-11 08:49:17 +00:00
return (BUS_PROBE_DEFAULT);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
t++;
}
style(9) - space after keywords.
2009-03-11 08:49:17 +00:00
return (ENXIO);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_attach(device_t dev)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
struct txp_softc *sc;
struct ifnet *ifp;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_rsp_desc *rsp;
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint16_t p1;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
uint32_t p2, reg;
int error = 0, pmc, rid;
uint8_t eaddr[ETHER_ADDR_LEN], *ver;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sc = device_get_softc(dev);
sc->sc_dev = dev;
Change callers of mtx_init() to pass in an appropriate lock type name. In most cases NULL is passed, but in some cases such as network driver locks (which use the MTX_NETWORK_LOCK macro) and UMA zone locks, a name is used. Tested on: i386, alpha, sparc64
2002-04-04 21:03:38 +00:00
mtx_init(&sc->sc_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
MTX_DEF);
callout_init_mtx(&sc->sc_tick, &sc->sc_mtx, 0);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TASK_INIT(&sc->sc_int_task, 0, txp_int_task, sc);
TAILQ_INIT(&sc->sc_busy_list);
TAILQ_INIT(&sc->sc_free_list);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ifmedia_init(&sc->sc_ifmedia, 0, txp_ifmedia_upd, txp_ifmedia_sts);
ifmedia_add(&sc->sc_ifmedia, IFM_ETHER | IFM_10_T, 0, NULL);
ifmedia_add(&sc->sc_ifmedia, IFM_ETHER | IFM_10_T | IFM_HDX, 0, NULL);
ifmedia_add(&sc->sc_ifmedia, IFM_ETHER | IFM_10_T | IFM_FDX, 0, NULL);
ifmedia_add(&sc->sc_ifmedia, IFM_ETHER | IFM_100_TX, 0, NULL);
ifmedia_add(&sc->sc_ifmedia, IFM_ETHER | IFM_100_TX | IFM_HDX, 0, NULL);
ifmedia_add(&sc->sc_ifmedia, IFM_ETHER | IFM_100_TX | IFM_FDX, 0, NULL);
ifmedia_add(&sc->sc_ifmedia, IFM_ETHER | IFM_AUTO, 0, NULL);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
pci_enable_busmaster(dev);
/* Prefer memory space register mapping over IO space. */
if (prefer_iomap == 0) {
sc->sc_res_id = PCIR_BAR(1);
sc->sc_res_type = SYS_RES_MEMORY;
} else {
sc->sc_res_id = PCIR_BAR(0);
sc->sc_res_type = SYS_RES_IOPORT;
}
sc->sc_res = bus_alloc_resource_any(dev, sc->sc_res_type,
&sc->sc_res_id, RF_ACTIVE);
if (sc->sc_res == NULL && prefer_iomap == 0) {
sc->sc_res_id = PCIR_BAR(0);
sc->sc_res_type = SYS_RES_IOPORT;
sc->sc_res = bus_alloc_resource_any(dev, sc->sc_res_type,
&sc->sc_res_id, RF_ACTIVE);
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (sc->sc_res == NULL) {
device_printf(dev, "couldn't map ports/memory\n");
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ifmedia_removeall(&sc->sc_ifmedia);
mtx_destroy(&sc->sc_mtx);
return (ENXIO);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Enable MWI. */
reg = pci_read_config(dev, PCIR_COMMAND, 2);
reg |= PCIM_CMD_MWRICEN;
pci_write_config(dev, PCIR_COMMAND, reg, 2);
/* Check cache line size. */
reg = pci_read_config(dev, PCIR_CACHELNSZ, 1);
reg <<= 4;
if (reg == 0 || (reg % 16) != 0)
device_printf(sc->sc_dev,
"invalid cache line size : %u\n", reg);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
/* Allocate interrupt */
rid = 0;
Convert callers to the new bus_alloc_resource_any(9) API. Submitted by: Mark Santcroos <marks@ripe.net> Reviewed by: imp, dfr, bde
2004-03-17 17:50:55 +00:00
sc->sc_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
RF_SHAREABLE | RF_ACTIVE);
if (sc->sc_irq == NULL) {
device_printf(dev, "couldn't map interrupt\n");
error = ENXIO;
goto fail;
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if ((error = txp_alloc_rings(sc)) != 0)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
goto fail;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_init_rings(sc);
txp_sysctl_node(sc);
/* Reset controller and make it reload sleep image. */
if (txp_reset(sc) != 0) {
Add checks for contigmalloc(9) failure.
2007-06-12 04:30:30 +00:00
error = ENXIO;
goto fail;
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Let controller boot from sleep image. */
if (txp_boot(sc, STAT_WAITING_FOR_HOST_REQUEST) != 0) {
device_printf(sc->sc_dev, "could not boot sleep image\n");
- Only call txp_release_resources() once if attach fails. - Set errno to ENXIO instead of 0 in several attach failure cases. - Setup the interrupt handler at the very end of txp_attach() after ether_ifattach(). - Various whitespace fixes in function prototypes.
2005-08-31 18:09:54 +00:00
error = ENXIO;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
goto fail;
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Get station address. */
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (txp_command(sc, TXP_CMD_STATION_ADDRESS_READ, 0, 0, 0,
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
&p1, &p2, NULL, TXP_CMD_WAIT)) {
- Only call txp_release_resources() once if attach fails. - Set errno to ENXIO instead of 0 in several attach failure cases. - Setup the interrupt handler at the very end of txp_attach() after ether_ifattach(). - Various whitespace fixes in function prototypes.
2005-08-31 18:09:54 +00:00
error = ENXIO;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
goto fail;
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
p1 = le16toh(p1);
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
eaddr[0] = ((uint8_t *)&p1)[1];
eaddr[1] = ((uint8_t *)&p1)[0];
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
p2 = le32toh(p2);
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
eaddr[2] = ((uint8_t *)&p2)[3];
eaddr[3] = ((uint8_t *)&p2)[2];
eaddr[4] = ((uint8_t *)&p2)[1];
eaddr[5] = ((uint8_t *)&p2)[0];
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
if (ifp == NULL) {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
device_printf(dev, "can not allocate ifnet structure\n");
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
error = ENOSPC;
goto fail;
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/*
* Show sleep image version information which may help to
* diagnose sleep image specific issues.
*/
rsp = NULL;
if (txp_ext_command(sc, TXP_CMD_READ_VERSION, 0, 0, 0, NULL, 0,
&rsp, TXP_CMD_WAIT)) {
device_printf(dev, "can not read sleep image version\n");
error = ENXIO;
goto fail;
}
if (rsp->rsp_numdesc == 0) {
p2 = le32toh(rsp->rsp_par2) & 0xFFFF;
device_printf(dev, "Typhoon 1.0 sleep image (2000/%02u/%02u)\n",
p2 >> 8, p2 & 0xFF);
} else if (rsp->rsp_numdesc == 2) {
p2 = le32toh(rsp->rsp_par2);
ver = (uint8_t *)(rsp + 1);
/*
* Even if datasheet says the command returns a NULL
* terminated version string, explicitly terminate
* the string. Given that several bugs of firmware
* I can't trust this simple one.
*/
ver[25] = '\0';
device_printf(dev,
"Typhoon 1.1+ sleep image %02u.%03u.%03u %s\n",
p2 >> 24, (p2 >> 12) & 0xFFF, p2 & 0xFFF, ver);
} else {
p2 = le32toh(rsp->rsp_par2);
device_printf(dev,
"Unknown Typhoon sleep image version: %u:0x%08x\n",
rsp->rsp_numdesc, p2);
}
if (rsp != NULL)
free(rsp, M_DEVBUF);
sc->sc_xcvr = TXP_XCVR_AUTO;
txp_command(sc, TXP_CMD_XCVR_SELECT, TXP_XCVR_AUTO, 0, 0,
NULL, NULL, NULL, TXP_CMD_NOWAIT);
ifmedia_set(&sc->sc_ifmedia, IFM_ETHER | IFM_AUTO);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
ifp->if_softc = sc;
Replace the if_name and if_unit members of struct ifnet with new members if_xname, if_dname, and if_dunit. if_xname is the name of the interface and if_dname/unit are the driver name and instance. This change paves the way for interface renaming and enhanced pseudo device creation and configuration symantics. Approved By: re (in principle) Reviewed By: njl, imp Tested On: i386, amd64, sparc64 Obtained From: NetBSD (if_xname)
2003-10-31 18:32:15 +00:00
if_initname(ifp, device_get_name(dev), device_get_unit(dev));
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
ifp->if_ioctl = txp_ioctl;
ifp->if_start = txp_start;
ifp->if_init = txp_init;
- Provide txp_get_counter() to return counters that are not collected, but taken from hardware. - Mechanically convert to if_inc_counter() the rest of counters.
2014-09-24 11:58:23 +00:00
ifp->if_get_counter = txp_get_counter;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ifp->if_snd.ifq_drv_maxlen = TX_ENTRIES - 1;
IFQ_SET_MAXLEN(&ifp->if_snd, ifp->if_snd.ifq_drv_maxlen);
IFQ_SET_READY(&ifp->if_snd);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
/*
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
* It's possible to read firmware's offload capability but
* we have not downloaded the firmware yet so announce
* working capability here. We're not interested in IPSec
* capability and due to the lots of firmware bug we can't
* advertise the whole capability anyway.
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
*/
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ifp->if_capabilities = IFCAP_RXCSUM | IFCAP_TXCSUM;
Do a sweep of the tree replacing calls to pci_find_extcap() with calls to pci_find_cap() instead.
2011-03-23 13:10:15 +00:00
if (pci_find_cap(dev, PCIY_PMG, &pmc) == 0)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ifp->if_capabilities |= IFCAP_WOL_MAGIC;
/* Enable all capabilities. */
ifp->if_capenable = ifp->if_capabilities;
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ether_ifattach(ifp, eaddr);
- Only call txp_release_resources() once if attach fails. - Set errno to ENXIO instead of 0 in several attach failure cases. - Setup the interrupt handler at the very end of txp_attach() after ether_ifattach(). - Various whitespace fixes in function prototypes.
2005-08-31 18:09:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* VLAN capability setup. */
ifp->if_capabilities |= IFCAP_VLAN_MTU;
ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
ifp->if_capenable = ifp->if_capabilities;
/* Tell the upper layer(s) we support long frames. */
Use define from if_var.h to access a field inside struct if_data, that resides in struct ifnet. Sponsored by: Nginx, Inc.
2014-08-30 19:55:54 +00:00
ifp->if_hdrlen = sizeof(struct ether_vlan_header);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
WRITE_REG(sc, TXP_IER, TXP_INTR_NONE);
WRITE_REG(sc, TXP_IMR, TXP_INTR_ALL);
/* Create local taskq. */
sc->sc_tq = taskqueue_create_fast("txp_taskq", M_WAITOK,
taskqueue_thread_enqueue, &sc->sc_tq);
if (sc->sc_tq == NULL) {
device_printf(dev, "could not create taskqueue.\n");
ether_ifdetach(ifp);
error = ENXIO;
goto fail;
}
taskqueue_start_threads(&sc->sc_tq, 1, PI_NET, "%s taskq",
device_get_nameunit(sc->sc_dev));
/* Put controller into sleep. */
if (txp_sleep(sc, 0) != 0) {
ether_ifdetach(ifp);
error = ENXIO;
goto fail;
}
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
error = bus_setup_intr(dev, sc->sc_irq, INTR_TYPE_NET | INTR_MPSAFE,
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_intr, NULL, sc, &sc->sc_intrhand);
- Only call txp_release_resources() once if attach fails. - Set errno to ENXIO instead of 0 in several attach failure cases. - Setup the interrupt handler at the very end of txp_attach() after ether_ifattach(). - Various whitespace fixes in function prototypes.
2005-08-31 18:09:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (error != 0) {
- Only call txp_release_resources() once if attach fails. - Set errno to ENXIO instead of 0 in several attach failure cases. - Setup the interrupt handler at the very end of txp_attach() after ether_ifattach(). - Various whitespace fixes in function prototypes.
2005-08-31 18:09:54 +00:00
ether_ifdetach(ifp);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
device_printf(dev, "couldn't set up interrupt handler.\n");
- Only call txp_release_resources() once if attach fails. - Set errno to ENXIO instead of 0 in several attach failure cases. - Setup the interrupt handler at the very end of txp_attach() after ether_ifattach(). - Various whitespace fixes in function prototypes.
2005-08-31 18:09:54 +00:00
goto fail;
}
style(9) - space after keywords.
2009-03-11 08:49:17 +00:00
return (0);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
fail:
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (error != 0)
txp_detach(dev);
style(9) - space after keywords.
2009-03-11 08:49:17 +00:00
return (error);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_detach(device_t dev)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
struct txp_softc *sc;
struct ifnet *ifp;
sc = device_get_softc(dev);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ifp = sc->sc_ifp;
if (device_is_attached(dev)) {
TXP_LOCK(sc);
sc->sc_flags |= TXP_FLAG_DETACH;
txp_stop(sc);
TXP_UNLOCK(sc);
callout_drain(&sc->sc_tick);
taskqueue_drain(sc->sc_tq, &sc->sc_int_task);
ether_ifdetach(ifp);
}
WRITE_REG(sc, TXP_IMR, TXP_INTR_ALL);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
ifmedia_removeall(&sc->sc_ifmedia);
if (sc->sc_intrhand != NULL)
bus_teardown_intr(dev, sc->sc_irq, sc->sc_intrhand);
if (sc->sc_irq != NULL)
bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq);
if (sc->sc_res != NULL)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_release_resource(dev, sc->sc_res_type, sc->sc_res_id,
sc->sc_res);
if (sc->sc_ifp != NULL) {
Make sure that we call if_free(ifp) after bus_teardown_intr. Since we could get an interrupt after we free the ifp, and the interrupt handler depended on the ifp being still alive, this could, in theory, cause a crash. Eliminate this possibility by moving the if_free to after the bus_teardown_intr() call.
2005-09-19 03:10:21 +00:00
if_free(sc->sc_ifp);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sc->sc_ifp = NULL;
}
txp_free_rings(sc);
mtx_destroy(&sc->sc_mtx);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return (0);
}
static int
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_reset(struct txp_softc *sc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint32_t r;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
int i;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Disable interrupts. */
WRITE_REG(sc, TXP_IER, TXP_INTR_NONE);
WRITE_REG(sc, TXP_IMR, TXP_INTR_ALL);
/* Ack all pending interrupts. */
WRITE_REG(sc, TXP_ISR, TXP_INTR_ALL);
Initialize variables before use. This was needed to pass -Werror. Reviewed by: jake
2002-03-19 05:14:23 +00:00
r = 0;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
WRITE_REG(sc, TXP_SRR, TXP_SRR_ALL);
DELAY(1000);
WRITE_REG(sc, TXP_SRR, 0);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Should wait max 6 seconds. */
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
for (i = 0; i < 6000; i++) {
r = READ_REG(sc, TXP_A2H_0);
if (r == STAT_WAITING_FOR_HOST_REQUEST)
break;
DELAY(1000);
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (r != STAT_WAITING_FOR_HOST_REQUEST)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
device_printf(sc->sc_dev, "reset hung\n");
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
WRITE_REG(sc, TXP_IER, TXP_INTR_NONE);
WRITE_REG(sc, TXP_IMR, TXP_INTR_ALL);
WRITE_REG(sc, TXP_ISR, TXP_INTR_ALL);
/*
* Give more time to complete loading sleep image before
* trying to boot from sleep image.
*/
DELAY(5000);
return (0);
}
static int
txp_boot(struct txp_softc *sc, uint32_t state)
{
/* See if it's waiting for boot, and try to boot it. */
if (txp_wait(sc, state) != 0) {
device_printf(sc->sc_dev, "not waiting for boot\n");
return (ENXIO);
}
WRITE_REG(sc, TXP_H2A_2, TXP_ADDR_HI(sc->sc_ldata.txp_boot_paddr));
TXP_BARRIER(sc, TXP_H2A_2, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_1, TXP_ADDR_LO(sc->sc_ldata.txp_boot_paddr));
TXP_BARRIER(sc, TXP_H2A_1, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_0, TXP_BOOTCMD_REGISTER_BOOT_RECORD);
TXP_BARRIER(sc, TXP_H2A_0, 4, BUS_SPACE_BARRIER_WRITE);
/* See if it booted. */
if (txp_wait(sc, STAT_RUNNING) != 0) {
device_printf(sc->sc_dev, "firmware not running\n");
return (ENXIO);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Clear TX and CMD ring write registers. */
WRITE_REG(sc, TXP_H2A_1, TXP_BOOTCMD_NULL);
TXP_BARRIER(sc, TXP_H2A_1, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_2, TXP_BOOTCMD_NULL);
TXP_BARRIER(sc, TXP_H2A_2, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_3, TXP_BOOTCMD_NULL);
TXP_BARRIER(sc, TXP_H2A_3, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_0, TXP_BOOTCMD_NULL);
TXP_BARRIER(sc, TXP_H2A_0, 4, BUS_SPACE_BARRIER_WRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return (0);
}
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_download_fw(struct txp_softc *sc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
struct txp_fw_file_header *fileheader;
struct txp_fw_section_header *secthead;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
int sect;
uint32_t error, ier, imr;
TXP_LOCK_ASSERT(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
Update to latest 3Com firmware image. The latest fimware is required to make 3CR990 familiy controllers run on NV flash firmware version 03.001.008. The latest firmware added HMAC digest information so teach txp(4) to pass them to sleep image before downloading is started. While I'm here restore previous IMR/IER register if firmware downloading have failed. PR: kern/89876, kern/132047
2009-02-25 01:12:56 +00:00
error = 0;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
ier = READ_REG(sc, TXP_IER);
WRITE_REG(sc, TXP_IER, ier | TXP_INT_A2H_0);
imr = READ_REG(sc, TXP_IMR);
WRITE_REG(sc, TXP_IMR, imr | TXP_INT_A2H_0);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (txp_wait(sc, STAT_WAITING_FOR_HOST_REQUEST) != 0) {
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
device_printf(sc->sc_dev, "not waiting for host request\n");
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
error = ETIMEDOUT;
Update to latest 3Com firmware image. The latest fimware is required to make 3CR990 familiy controllers run on NV flash firmware version 03.001.008. The latest firmware added HMAC digest information so teach txp(4) to pass them to sleep image before downloading is started. While I'm here restore previous IMR/IER register if firmware downloading have failed. PR: kern/89876, kern/132047
2009-02-25 01:12:56 +00:00
goto fail;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Ack the status. */
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
WRITE_REG(sc, TXP_ISR, TXP_INT_A2H_0);
fileheader = (struct txp_fw_file_header *)tc990image;
if (bcmp("TYPHOON", fileheader->magicid, sizeof(fileheader->magicid))) {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
device_printf(sc->sc_dev, "firmware invalid magic\n");
Update to latest 3Com firmware image. The latest fimware is required to make 3CR990 familiy controllers run on NV flash firmware version 03.001.008. The latest firmware added HMAC digest information so teach txp(4) to pass them to sleep image before downloading is started. While I'm here restore previous IMR/IER register if firmware downloading have failed. PR: kern/89876, kern/132047
2009-02-25 01:12:56 +00:00
goto fail;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Tell boot firmware to get ready for image. */
WRITE_REG(sc, TXP_H2A_1, le32toh(fileheader->addr));
TXP_BARRIER(sc, TXP_H2A_1, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_2, le32toh(fileheader->hmac[0]));
TXP_BARRIER(sc, TXP_H2A_2, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_3, le32toh(fileheader->hmac[1]));
TXP_BARRIER(sc, TXP_H2A_3, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_4, le32toh(fileheader->hmac[2]));
TXP_BARRIER(sc, TXP_H2A_4, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_5, le32toh(fileheader->hmac[3]));
TXP_BARRIER(sc, TXP_H2A_5, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_6, le32toh(fileheader->hmac[4]));
TXP_BARRIER(sc, TXP_H2A_6, 4, BUS_SPACE_BARRIER_WRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
WRITE_REG(sc, TXP_H2A_0, TXP_BOOTCMD_RUNTIME_IMAGE);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_BARRIER(sc, TXP_H2A_0, 4, BUS_SPACE_BARRIER_WRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (txp_download_fw_wait(sc)) {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
device_printf(sc->sc_dev, "firmware wait failed, initial\n");
error = ETIMEDOUT;
Update to latest 3Com firmware image. The latest fimware is required to make 3CR990 familiy controllers run on NV flash firmware version 03.001.008. The latest firmware added HMAC digest information so teach txp(4) to pass them to sleep image before downloading is started. While I'm here restore previous IMR/IER register if firmware downloading have failed. PR: kern/89876, kern/132047
2009-02-25 01:12:56 +00:00
goto fail;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
secthead = (struct txp_fw_section_header *)(((uint8_t *)tc990image) +
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sizeof(struct txp_fw_file_header));
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
for (sect = 0; sect < le32toh(fileheader->nsections); sect++) {
if ((error = txp_download_fw_section(sc, secthead, sect)) != 0)
Update to latest 3Com firmware image. The latest fimware is required to make 3CR990 familiy controllers run on NV flash firmware version 03.001.008. The latest firmware added HMAC digest information so teach txp(4) to pass them to sleep image before downloading is started. While I'm here restore previous IMR/IER register if firmware downloading have failed. PR: kern/89876, kern/132047
2009-02-25 01:12:56 +00:00
goto fail;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
secthead = (struct txp_fw_section_header *)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
(((uint8_t *)secthead) + le32toh(secthead->nbytes) +
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sizeof(*secthead));
}
WRITE_REG(sc, TXP_H2A_0, TXP_BOOTCMD_DOWNLOAD_COMPLETE);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_BARRIER(sc, TXP_H2A_0, 4, BUS_SPACE_BARRIER_WRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (txp_wait(sc, STAT_WAITING_FOR_BOOT) != 0) {
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
device_printf(sc->sc_dev, "not waiting for boot\n");
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
error = ETIMEDOUT;
Update to latest 3Com firmware image. The latest fimware is required to make 3CR990 familiy controllers run on NV flash firmware version 03.001.008. The latest firmware added HMAC digest information so teach txp(4) to pass them to sleep image before downloading is started. While I'm here restore previous IMR/IER register if firmware downloading have failed. PR: kern/89876, kern/132047
2009-02-25 01:12:56 +00:00
goto fail;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
Update to latest 3Com firmware image. The latest fimware is required to make 3CR990 familiy controllers run on NV flash firmware version 03.001.008. The latest firmware added HMAC digest information so teach txp(4) to pass them to sleep image before downloading is started. While I'm here restore previous IMR/IER register if firmware downloading have failed. PR: kern/89876, kern/132047
2009-02-25 01:12:56 +00:00
fail:
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
WRITE_REG(sc, TXP_IER, ier);
WRITE_REG(sc, TXP_IMR, imr);
Update to latest 3Com firmware image. The latest fimware is required to make 3CR990 familiy controllers run on NV flash firmware version 03.001.008. The latest firmware added HMAC digest information so teach txp(4) to pass them to sleep image before downloading is started. While I'm here restore previous IMR/IER register if firmware downloading have failed. PR: kern/89876, kern/132047
2009-02-25 01:12:56 +00:00
return (error);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_download_fw_wait(struct txp_softc *sc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
uint32_t i;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_LOCK_ASSERT(sc);
for (i = 0; i < TXP_TIMEOUT; i++) {
if ((READ_REG(sc, TXP_ISR) & TXP_INT_A2H_0) != 0)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
DELAY(50);
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (i == TXP_TIMEOUT) {
device_printf(sc->sc_dev, "firmware wait failed comm0\n");
return (ETIMEDOUT);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
WRITE_REG(sc, TXP_ISR, TXP_INT_A2H_0);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (READ_REG(sc, TXP_A2H_0) != STAT_WAITING_FOR_SEGMENT) {
device_printf(sc->sc_dev, "firmware not waiting for segment\n");
return (ETIMEDOUT);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
return (0);
}
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_download_fw_section(struct txp_softc *sc,
struct txp_fw_section_header *sect, int sectnum)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dma_tag_t sec_tag;
bus_dmamap_t sec_map;
bus_addr_t sec_paddr;
uint8_t *sec_buf;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
int rseg, err = 0;
struct mbuf m;
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint16_t csum;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_LOCK_ASSERT(sc);
/* Skip zero length sections. */
if (le32toh(sect->nbytes) == 0)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return (0);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Make sure we aren't past the end of the image. */
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
rseg = ((uint8_t *)sect) - ((uint8_t *)tc990image);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (rseg >= sizeof(tc990image)) {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
device_printf(sc->sc_dev,
"firmware invalid section address, section %d\n", sectnum);
return (EIO);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Make sure this section doesn't go past the end. */
rseg += le32toh(sect->nbytes);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (rseg >= sizeof(tc990image)) {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
device_printf(sc->sc_dev, "firmware truncated section %d\n",
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sectnum);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
return (EIO);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sec_tag = NULL;
sec_map = NULL;
sec_buf = NULL;
/* XXX */
TXP_UNLOCK(sc);
err = txp_dma_alloc(sc, "firmware sections", &sec_tag, sizeof(uint32_t),
0, &sec_map, (void **)&sec_buf, le32toh(sect->nbytes), &sec_paddr);
TXP_LOCK(sc);
if (err != 0)
goto bail;
bcopy(((uint8_t *)sect) + sizeof(*sect), sec_buf,
le32toh(sect->nbytes));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
/*
* dummy up mbuf and verify section checksum
*/
m.m_type = MT_DATA;
m.m_next = m.m_nextpkt = NULL;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
m.m_len = le32toh(sect->nbytes);
m.m_data = sec_buf;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
m.m_flags = 0;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
csum = in_cksum(&m, le32toh(sect->nbytes));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (csum != sect->cksum) {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
device_printf(sc->sc_dev,
"firmware section %d, bad cksum (expected 0x%x got 0x%x)\n",
sectnum, le16toh(sect->cksum), csum);
err = EIO;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
goto bail;
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sec_tag, sec_map, BUS_DMASYNC_PREWRITE);
WRITE_REG(sc, TXP_H2A_1, le32toh(sect->nbytes));
TXP_BARRIER(sc, TXP_H2A_1, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_2, le16toh(sect->cksum));
TXP_BARRIER(sc, TXP_H2A_2, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_3, le32toh(sect->addr));
TXP_BARRIER(sc, TXP_H2A_3, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_4, TXP_ADDR_HI(sec_paddr));
TXP_BARRIER(sc, TXP_H2A_4, 4, BUS_SPACE_BARRIER_WRITE);
WRITE_REG(sc, TXP_H2A_5, TXP_ADDR_LO(sec_paddr));
TXP_BARRIER(sc, TXP_H2A_5, 4, BUS_SPACE_BARRIER_WRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
WRITE_REG(sc, TXP_H2A_0, TXP_BOOTCMD_SEGMENT_AVAILABLE);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_BARRIER(sc, TXP_H2A_0, 4, BUS_SPACE_BARRIER_WRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (txp_download_fw_wait(sc)) {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
device_printf(sc->sc_dev,
"firmware wait failed, section %d\n", sectnum);
err = ETIMEDOUT;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sec_tag, sec_map, BUS_DMASYNC_POSTWRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bail:
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
txp_dma_free(sc, &sec_tag, sec_map, (void **)&sec_buf, &sec_paddr);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return (err);
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_intr(void *vsc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_softc *sc;
uint32_t status;
sc = vsc;
status = READ_REG(sc, TXP_ISR);
if ((status & TXP_INT_LATCH) == 0)
return (FILTER_STRAY);
WRITE_REG(sc, TXP_ISR, status);
WRITE_REG(sc, TXP_IMR, TXP_INTR_ALL);
taskqueue_enqueue(sc->sc_tq, &sc->sc_int_task);
return (FILTER_HANDLED);
}
static void
txp_int_task(void *arg, int pending)
{
struct txp_softc *sc;
struct ifnet *ifp;
struct txp_hostvar *hv;
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint32_t isr;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
int more;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sc = (struct txp_softc *)arg;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_LOCK(sc);
ifp = sc->sc_ifp;
hv = sc->sc_hostvar;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
isr = READ_REG(sc, TXP_ISR);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if ((isr & TXP_INT_LATCH) != 0)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
WRITE_REG(sc, TXP_ISR, isr);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
bus_dmamap_sync(sc->sc_cdata.txp_hostvar_tag,
sc->sc_cdata.txp_hostvar_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
more = 0;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if ((*sc->sc_rxhir.r_roff) != (*sc->sc_rxhir.r_woff))
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
more += txp_rx_reclaim(sc, &sc->sc_rxhir,
sc->sc_process_limit);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if ((*sc->sc_rxlor.r_roff) != (*sc->sc_rxlor.r_woff))
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
more += txp_rx_reclaim(sc, &sc->sc_rxlor,
sc->sc_process_limit);
/*
* XXX
* It seems controller is not smart enough to handle
* FIFO overflow conditions under heavy network load.
* No matter how often new Rx buffers are passed to
* controller the situation didn't change. Maybe
* flow-control would be the only way to mitigate the
* issue but firmware does not have commands that
* control the threshold of emitting pause frames.
*/
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (hv->hv_rx_buf_write_idx == hv->hv_rx_buf_read_idx)
txp_rxbuf_reclaim(sc);
if (sc->sc_txhir.r_cnt && (sc->sc_txhir.r_cons !=
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_OFFSET2IDX(le32toh(*(sc->sc_txhir.r_off)))))
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
txp_tx_reclaim(sc, &sc->sc_txhir);
if (sc->sc_txlor.r_cnt && (sc->sc_txlor.r_cons !=
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_OFFSET2IDX(le32toh(*(sc->sc_txlor.r_off)))))
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
txp_tx_reclaim(sc, &sc->sc_txlor);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sc->sc_cdata.txp_hostvar_tag,
sc->sc_cdata.txp_hostvar_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
txp_start_locked(sc->sc_ifp);
if (more != 0 || READ_REG(sc, TXP_ISR & TXP_INT_LATCH) != 0) {
taskqueue_enqueue(sc->sc_tq, &sc->sc_int_task);
TXP_UNLOCK(sc);
return;
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Re-enable interrupts. */
WRITE_REG(sc, TXP_IMR, TXP_INTR_NONE);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_UNLOCK(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#ifndef __NO_STRICT_ALIGNMENT
static __inline void
txp_fixup_rx(struct mbuf *m)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
int i;
uint16_t *src, *dst;
src = mtod(m, uint16_t *);
dst = src - (TXP_RXBUF_ALIGN - ETHER_ALIGN) / sizeof *src;
for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
*dst++ = *src++;
m->m_data -= TXP_RXBUF_ALIGN - ETHER_ALIGN;
}
#endif
static int
txp_rx_reclaim(struct txp_softc *sc, struct txp_rx_ring *r, int count)
{
struct ifnet *ifp;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
struct txp_rx_desc *rxd;
struct mbuf *m;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_rx_swdesc *sd;
uint32_t roff, woff, rx_stat, prog;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK_ASSERT(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ifp = sc->sc_ifp;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(r->r_tag, r->r_map, BUS_DMASYNC_POSTREAD |
BUS_DMASYNC_POSTWRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
roff = le32toh(*r->r_roff);
woff = le32toh(*r->r_woff);
rxd = r->r_desc + roff / sizeof(struct txp_rx_desc);
for (prog = 0; roff != woff; prog++, count--) {
if (count <= 0)
break;
bcopy((u_long *)&rxd->rx_vaddrlo, &sd, sizeof(sd));
KASSERT(sd != NULL, ("%s: Rx desc ring corrupted", __func__));
bus_dmamap_sync(sc->sc_cdata.txp_rx_tag, sd->sd_map,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->sc_cdata.txp_rx_tag, sd->sd_map);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
m = sd->sd_mbuf;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
KASSERT(m != NULL, ("%s: Rx buffer ring corrupted", __func__));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sd->sd_mbuf = NULL;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TAILQ_REMOVE(&sc->sc_busy_list, sd, sd_next);
TAILQ_INSERT_TAIL(&sc->sc_free_list, sd, sd_next);
if ((rxd->rx_flags & RX_FLAGS_ERROR) != 0) {
if (bootverbose)
device_printf(sc->sc_dev, "Rx error %u\n",
le32toh(rxd->rx_stat) & RX_ERROR_MASK);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
m_freem(m);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
goto next;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
m->m_pkthdr.len = m->m_len = le16toh(rxd->rx_len);
m->m_pkthdr.rcvif = ifp;
#ifndef __NO_STRICT_ALIGNMENT
txp_fixup_rx(m);
#endif
rx_stat = le32toh(rxd->rx_stat);
if ((ifp->if_capenable & IFCAP_RXCSUM) != 0) {
if ((rx_stat & RX_STAT_IPCKSUMBAD) != 0)
m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
else if ((rx_stat & RX_STAT_IPCKSUMGOOD) != 0)
m->m_pkthdr.csum_flags |=
CSUM_IP_CHECKED|CSUM_IP_VALID;
if ((rx_stat & RX_STAT_TCPCKSUMGOOD) != 0 ||
(rx_stat & RX_STAT_UDPCKSUMGOOD) != 0) {
m->m_pkthdr.csum_flags |=
CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
m->m_pkthdr.csum_data = 0xffff;
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/*
* XXX
* Typhoon has a firmware bug that VLAN tag is always
* stripped out even if it is told to not remove the tag.
* Therefore don't check if_capenable here.
*/
if (/* (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 && */
(rx_stat & RX_STAT_VLAN) != 0) {
m->m_pkthdr.ether_vtag =
bswap16((le32toh(rxd->rx_vlan) >> 16));
Move ethernet VLAN tags from mtags to its own mbuf packet header field m_pkthdr.ether_vlan. The presence of the M_VLANTAG flag on the mbuf signifies the presence and validity of its content. Drivers that support hardware VLAN tag stripping fill in the received VLAN tag (containing both vlan and priority information) into the ether_vtag mbuf packet header field: m->m_pkthdr.ether_vtag = vlan_id; /* ntohs()? */ m->m_flags |= M_VLANTAG; to mark the packet m with the specified VLAN tag. On output the driver should check the mbuf for the M_VLANTAG flag to see if a VLAN tag is present and valid: if (m->m_flags & M_VLANTAG) { ... = m->m_pkthdr.ether_vtag; /* htons()? */ ... pass tag to hardware ... } VLAN tags are stored in host byte order. Byte swapping may be necessary. (Note: This driver conversion was mechanic and did not add or remove any byte swapping in the drivers.) Remove zone_mtag_vlan UMA zone and MTAG_VLAN definition. No more tag memory allocation have to be done. Reviewed by: thompsa, yar Sponsored by: TCP/IP Optimization Fundraise 2005
2006-09-17 13:33:30 +00:00
m->m_flags |= M_VLANTAG;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
Make vlan(4) loadable, unloadable, and clonable. As a side effect, interfaces must now always enable VLAN support. Reviewed by: jlemon MFC after: 3 weeks
2001-09-05 21:10:28 +00:00
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_UNLOCK(sc);
network interface driver changes: o don't strip the Ethernet header from inbound packets; pass packets up the stack intact (required significant changes to some drivers) o reference common definitions in net/ethernet.h (e.g. ETHER_ALIGN) o track ether_ifattach/ether_ifdetach API changes o track bpf changes (use BPF_TAP and BPF_MTAP) o track vlan changes (ifnet capabilities, revised processing scheme, etc.) o use if_input to pass packets "up" o call ether_ioctl for default handling of ioctls Reviewed by: many Approved by: re
2002-11-14 23:54:55 +00:00
(*ifp->if_input)(ifp, m);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
next:
roff += sizeof(struct txp_rx_desc);
if (roff == (RX_ENTRIES * sizeof(struct txp_rx_desc))) {
roff = 0;
rxd = r->r_desc;
} else
rxd++;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
prog++;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (prog == 0)
return (0);
bus_dmamap_sync(r->r_tag, r->r_map, BUS_DMASYNC_PREREAD |
BUS_DMASYNC_PREWRITE);
*r->r_roff = le32toh(roff);
return (count > 0 ? 0 : EAGAIN);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_rxbuf_reclaim(struct txp_softc *sc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_hostvar *hv;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
struct txp_rxbuf_desc *rbd;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_rx_swdesc *sd;
bus_dma_segment_t segs[1];
int nsegs, prod, prog;
uint32_t cons;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK_ASSERT(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
hv = sc->sc_hostvar;
cons = TXP_OFFSET2IDX(le32toh(hv->hv_rx_buf_read_idx));
prod = sc->sc_rxbufprod;
TXP_DESC_INC(prod, RXBUF_ENTRIES);
if (prod == cons)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sc->sc_cdata.txp_rxbufs_tag,
sc->sc_cdata.txp_rxbufs_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
for (prog = 0; prod != cons; prog++) {
sd = TAILQ_FIRST(&sc->sc_free_list);
if (sd == NULL)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
rbd = sc->sc_rxbufs + prod;
bcopy((u_long *)&rbd->rb_vaddrlo, &sd, sizeof(sd));
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
sd->sd_mbuf = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (sd->sd_mbuf == NULL)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
break;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sd->sd_mbuf->m_pkthdr.len = sd->sd_mbuf->m_len = MCLBYTES;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#ifndef __NO_STRICT_ALIGNMENT
m_adj(sd->sd_mbuf, TXP_RXBUF_ALIGN);
#endif
if (bus_dmamap_load_mbuf_sg(sc->sc_cdata.txp_rx_tag,
sd->sd_map, sd->sd_mbuf, segs, &nsegs, 0) != 0) {
m_freem(sd->sd_mbuf);
sd->sd_mbuf = NULL;
break;
}
KASSERT(nsegs == 1, ("%s : %d segments returned!", __func__,
nsegs));
TAILQ_REMOVE(&sc->sc_free_list, sd, sd_next);
TAILQ_INSERT_TAIL(&sc->sc_busy_list, sd, sd_next);
bus_dmamap_sync(sc->sc_cdata.txp_rx_tag, sd->sd_map,
BUS_DMASYNC_PREREAD);
rbd->rb_paddrlo = htole32(TXP_ADDR_LO(segs[0].ds_addr));
rbd->rb_paddrhi = htole32(TXP_ADDR_HI(segs[0].ds_addr));
TXP_DESC_INC(prod, RXBUF_ENTRIES);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (prog == 0)
return;
bus_dmamap_sync(sc->sc_cdata.txp_rxbufs_tag,
sc->sc_cdata.txp_rxbufs_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
prod = (prod + RXBUF_ENTRIES - 1) % RXBUF_ENTRIES;
sc->sc_rxbufprod = prod;
hv->hv_rx_buf_write_idx = htole32(TXP_IDX2OFFSET(prod));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
/*
* Reclaim mbufs and entries from a transmit ring.
*/
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_tx_reclaim(struct txp_softc *sc, struct txp_tx_ring *r)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct ifnet *ifp;
uint32_t idx;
uint32_t cons, cnt;
struct txp_tx_desc *txd;
struct txp_swdesc *sd;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK_ASSERT(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(r->r_tag, r->r_map, BUS_DMASYNC_POSTREAD |
BUS_DMASYNC_POSTWRITE);
ifp = sc->sc_ifp;
idx = TXP_OFFSET2IDX(le32toh(*(r->r_off)));
cons = r->r_cons;
cnt = r->r_cnt;
txd = r->r_desc + cons;
sd = sc->sc_txd + cons;
for (cnt = r->r_cnt; cons != idx && cnt > 0; cnt--) {
if ((txd->tx_flags & TX_FLAGS_TYPE_M) == TX_FLAGS_TYPE_DATA) {
if (sd->sd_mbuf != NULL) {
bus_dmamap_sync(sc->sc_cdata.txp_tx_tag,
sd->sd_map, BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(sc->sc_cdata.txp_tx_tag,
sd->sd_map);
m_freem(sd->sd_mbuf);
sd->sd_mbuf = NULL;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
txd->tx_addrlo = 0;
txd->tx_addrhi = 0;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txd->tx_flags = 0;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
}
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (++cons == TX_ENTRIES) {
txd = r->r_desc;
cons = 0;
sd = sc->sc_txd;
} else {
txd++;
sd++;
}
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(r->r_tag, r->r_map, BUS_DMASYNC_PREREAD |
BUS_DMASYNC_PREWRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
r->r_cons = cons;
r->r_cnt = cnt;
if (cnt == 0)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sc->sc_watchdog_timer = 0;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_shutdown(device_t dev)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
{
return (txp_suspend(dev));
}
static int
txp_suspend(device_t dev)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
struct txp_softc *sc;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct ifnet *ifp;
uint8_t *eaddr;
uint16_t p1;
uint32_t p2;
int pmc;
uint16_t pmstat;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sc = device_get_softc(dev);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ifp = sc->sc_ifp;
txp_stop(sc);
txp_init_rings(sc);
/* Reset controller and make it reload sleep image. */
txp_reset(sc);
/* Let controller boot from sleep image. */
if (txp_boot(sc, STAT_WAITING_FOR_HOST_REQUEST) != 0)
device_printf(sc->sc_dev, "couldn't boot sleep image\n");
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Set station address. */
eaddr = IF_LLADDR(sc->sc_ifp);
p1 = 0;
((uint8_t *)&p1)[1] = eaddr[0];
((uint8_t *)&p1)[0] = eaddr[1];
p1 = le16toh(p1);
((uint8_t *)&p2)[3] = eaddr[2];
((uint8_t *)&p2)[2] = eaddr[3];
((uint8_t *)&p2)[1] = eaddr[4];
((uint8_t *)&p2)[0] = eaddr[5];
p2 = le32toh(p2);
txp_command(sc, TXP_CMD_STATION_ADDRESS_WRITE, p1, p2, 0, NULL, NULL,
NULL, TXP_CMD_WAIT);
txp_set_filter(sc);
WRITE_REG(sc, TXP_IER, TXP_INTR_NONE);
WRITE_REG(sc, TXP_IMR, TXP_INTR_ALL);
txp_sleep(sc, sc->sc_ifp->if_capenable);
Do a sweep of the tree replacing calls to pci_find_extcap() with calls to pci_find_cap() instead.
2011-03-23 13:10:15 +00:00
if (pci_find_cap(sc->sc_dev, PCIY_PMG, &pmc) == 0) {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Request PME. */
pmstat = pci_read_config(sc->sc_dev,
pmc + PCIR_POWER_STATUS, 2);
pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
if ((ifp->if_capenable & IFCAP_WOL) != 0)
pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
pci_write_config(sc->sc_dev,
pmc + PCIR_POWER_STATUS, pmstat, 2);
}
TXP_UNLOCK(sc);
return (0);
}
static int
txp_resume(device_t dev)
{
struct txp_softc *sc;
int pmc;
uint16_t pmstat;
sc = device_get_softc(dev);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_LOCK(sc);
Do a sweep of the tree replacing calls to pci_find_extcap() with calls to pci_find_cap() instead.
2011-03-23 13:10:15 +00:00
if (pci_find_cap(sc->sc_dev, PCIY_PMG, &pmc) == 0) {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Disable PME and clear PME status. */
pmstat = pci_read_config(sc->sc_dev,
pmc + PCIR_POWER_STATUS, 2);
if ((pmstat & PCIM_PSTAT_PMEENABLE) != 0) {
pmstat &= ~PCIM_PSTAT_PMEENABLE;
pci_write_config(sc->sc_dev,
pmc + PCIR_POWER_STATUS, pmstat, 2);
}
}
if ((sc->sc_ifp->if_flags & IFF_UP) != 0)
txp_init_locked(sc);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_UNLOCK(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
style(9) - space after keywords.
2009-03-11 08:49:17 +00:00
return (0);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_dmamap_arg {
bus_addr_t txp_busaddr;
};
static void
txp_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
{
struct txp_dmamap_arg *ctx;
if (error != 0)
return;
KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
ctx = (struct txp_dmamap_arg *)arg;
ctx->txp_busaddr = segs[0].ds_addr;
}
static int
txp_dma_alloc(struct txp_softc *sc, char *type, bus_dma_tag_t *tag,
bus_size_t alignment, bus_size_t boundary, bus_dmamap_t *map, void **buf,
bus_size_t size, bus_addr_t *paddr)
{
struct txp_dmamap_arg ctx;
int error;
/* Create DMA block tag. */
error = bus_dma_tag_create(
sc->sc_cdata.txp_parent_tag, /* parent */
alignment, boundary, /* algnmnt, boundary */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
size, /* maxsize */
1, /* nsegments */
size, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
tag);
if (error != 0) {
device_printf(sc->sc_dev,
"could not create DMA tag for %s.\n", type);
return (error);
}
*paddr = 0;
/* Allocate DMA'able memory and load the DMA map. */
error = bus_dmamem_alloc(*tag, buf, BUS_DMA_WAITOK | BUS_DMA_ZERO |
BUS_DMA_COHERENT, map);
if (error != 0) {
device_printf(sc->sc_dev,
"could not allocate DMA'able memory for %s.\n", type);
return (error);
}
ctx.txp_busaddr = 0;
error = bus_dmamap_load(*tag, *map, *(uint8_t **)buf,
size, txp_dmamap_cb, &ctx, BUS_DMA_NOWAIT);
if (error != 0 || ctx.txp_busaddr == 0) {
device_printf(sc->sc_dev,
"could not load DMA'able memory for %s.\n", type);
return (error);
}
*paddr = ctx.txp_busaddr;
return (0);
}
static void
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
txp_dma_free(struct txp_softc *sc, bus_dma_tag_t *tag, bus_dmamap_t map,
void **buf, bus_addr_t *paddr)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
{
if (*tag != NULL) {
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
if (*paddr != 0)
bus_dmamap_unload(*tag, map);
if (buf != NULL)
bus_dmamem_free(*tag, *(uint8_t **)buf, map);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
*(uint8_t **)buf = NULL;
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
*paddr = 0;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dma_tag_destroy(*tag);
*tag = NULL;
}
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_alloc_rings(struct txp_softc *sc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
struct txp_boot_record *boot;
struct txp_ldata *ld;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_swdesc *txd;
struct txp_rxbuf_desc *rbd;
struct txp_rx_swdesc *sd;
int error, i;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ld = &sc->sc_ldata;
boot = ld->txp_boot;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
/* boot record */
sc->sc_boot = boot;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/*
* Create parent ring/DMA block tag.
* Datasheet says that all ring addresses and descriptors
* support 64bits addressing. However the controller is
* known to have no support DAC so limit DMA address space
* to 32bits.
*/
error = bus_dma_tag_create(
bus_get_dma_tag(sc->sc_dev), /* parent */
1, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR_32BIT, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
BUS_SPACE_MAXSIZE_32BIT, /* maxsize */
0, /* nsegments */
BUS_SPACE_MAXSIZE_32BIT, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->sc_cdata.txp_parent_tag);
if (error != 0) {
device_printf(sc->sc_dev, "could not create parent DMA tag.\n");
return (error);
}
/* Boot record. */
error = txp_dma_alloc(sc, "boot record",
&sc->sc_cdata.txp_boot_tag, sizeof(uint32_t), 0,
&sc->sc_cdata.txp_boot_map, (void **)&sc->sc_ldata.txp_boot,
sizeof(struct txp_boot_record),
&sc->sc_ldata.txp_boot_paddr);
if (error != 0)
return (error);
boot = sc->sc_ldata.txp_boot;
sc->sc_boot = boot;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Host variables. */
error = txp_dma_alloc(sc, "host variables",
&sc->sc_cdata.txp_hostvar_tag, sizeof(uint32_t), 0,
&sc->sc_cdata.txp_hostvar_map, (void **)&sc->sc_ldata.txp_hostvar,
sizeof(struct txp_hostvar),
&sc->sc_ldata.txp_hostvar_paddr);
if (error != 0)
return (error);
boot->br_hostvar_lo =
htole32(TXP_ADDR_LO(sc->sc_ldata.txp_hostvar_paddr));
boot->br_hostvar_hi =
htole32(TXP_ADDR_HI(sc->sc_ldata.txp_hostvar_paddr));
sc->sc_hostvar = sc->sc_ldata.txp_hostvar;
/* Hi priority tx ring. */
error = txp_dma_alloc(sc, "hi priority tx ring",
&sc->sc_cdata.txp_txhiring_tag, sizeof(struct txp_tx_desc), 0,
&sc->sc_cdata.txp_txhiring_map, (void **)&sc->sc_ldata.txp_txhiring,
sizeof(struct txp_tx_desc) * TX_ENTRIES,
&sc->sc_ldata.txp_txhiring_paddr);
if (error != 0)
return (error);
boot->br_txhipri_lo =
htole32(TXP_ADDR_LO(sc->sc_ldata.txp_txhiring_paddr));
boot->br_txhipri_hi =
htole32(TXP_ADDR_HI(sc->sc_ldata.txp_txhiring_paddr));
boot->br_txhipri_siz =
htole32(TX_ENTRIES * sizeof(struct txp_tx_desc));
sc->sc_txhir.r_tag = sc->sc_cdata.txp_txhiring_tag;
sc->sc_txhir.r_map = sc->sc_cdata.txp_txhiring_map;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sc->sc_txhir.r_reg = TXP_H2A_1;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sc->sc_txhir.r_desc = sc->sc_ldata.txp_txhiring;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sc->sc_txhir.r_cons = sc->sc_txhir.r_prod = sc->sc_txhir.r_cnt = 0;
sc->sc_txhir.r_off = &sc->sc_hostvar->hv_tx_hi_desc_read_idx;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Low priority tx ring. */
error = txp_dma_alloc(sc, "low priority tx ring",
&sc->sc_cdata.txp_txloring_tag, sizeof(struct txp_tx_desc), 0,
&sc->sc_cdata.txp_txloring_map, (void **)&sc->sc_ldata.txp_txloring,
sizeof(struct txp_tx_desc) * TX_ENTRIES,
&sc->sc_ldata.txp_txloring_paddr);
if (error != 0)
return (error);
boot->br_txlopri_lo =
htole32(TXP_ADDR_LO(sc->sc_ldata.txp_txloring_paddr));
boot->br_txlopri_hi =
htole32(TXP_ADDR_HI(sc->sc_ldata.txp_txloring_paddr));
boot->br_txlopri_siz =
htole32(TX_ENTRIES * sizeof(struct txp_tx_desc));
sc->sc_txlor.r_tag = sc->sc_cdata.txp_txloring_tag;
sc->sc_txlor.r_map = sc->sc_cdata.txp_txloring_map;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sc->sc_txlor.r_reg = TXP_H2A_3;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sc->sc_txlor.r_desc = sc->sc_ldata.txp_txloring;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sc->sc_txlor.r_cons = sc->sc_txlor.r_prod = sc->sc_txlor.r_cnt = 0;
sc->sc_txlor.r_off = &sc->sc_hostvar->hv_tx_lo_desc_read_idx;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* High priority rx ring. */
error = txp_dma_alloc(sc, "hi priority rx ring",
Make sure rx descriptor ring align on 16 bytes. I guess the alignment requirement could be multiple of 4 bytes but I think using descriptor size would make intention clearer. Previously the size of rx descriptor was not power of 2 so it caused panic in bus_dmamem_alloc(9). Reported by: Jeff Blank (jb000003 <> mr-happy dot com) MFC after: 3 days
2009-08-31 22:09:48 +00:00
&sc->sc_cdata.txp_rxhiring_tag,
roundup(sizeof(struct txp_rx_desc), 16), 0,
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
&sc->sc_cdata.txp_rxhiring_map, (void **)&sc->sc_ldata.txp_rxhiring,
sizeof(struct txp_rx_desc) * RX_ENTRIES,
&sc->sc_ldata.txp_rxhiring_paddr);
if (error != 0)
return (error);
boot->br_rxhipri_lo =
htole32(TXP_ADDR_LO(sc->sc_ldata.txp_rxhiring_paddr));
boot->br_rxhipri_hi =
htole32(TXP_ADDR_HI(sc->sc_ldata.txp_rxhiring_paddr));
boot->br_rxhipri_siz =
htole32(RX_ENTRIES * sizeof(struct txp_rx_desc));
sc->sc_rxhir.r_tag = sc->sc_cdata.txp_rxhiring_tag;
sc->sc_rxhir.r_map = sc->sc_cdata.txp_rxhiring_map;
sc->sc_rxhir.r_desc = sc->sc_ldata.txp_rxhiring;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sc->sc_rxhir.r_roff = &sc->sc_hostvar->hv_rx_hi_read_idx;
sc->sc_rxhir.r_woff = &sc->sc_hostvar->hv_rx_hi_write_idx;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Low priority rx ring. */
error = txp_dma_alloc(sc, "low priority rx ring",
Make sure rx descriptor ring align on 16 bytes. I guess the alignment requirement could be multiple of 4 bytes but I think using descriptor size would make intention clearer. Previously the size of rx descriptor was not power of 2 so it caused panic in bus_dmamem_alloc(9). Reported by: Jeff Blank (jb000003 <> mr-happy dot com) MFC after: 3 days
2009-08-31 22:09:48 +00:00
&sc->sc_cdata.txp_rxloring_tag,
roundup(sizeof(struct txp_rx_desc), 16), 0,
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
&sc->sc_cdata.txp_rxloring_map, (void **)&sc->sc_ldata.txp_rxloring,
sizeof(struct txp_rx_desc) * RX_ENTRIES,
&sc->sc_ldata.txp_rxloring_paddr);
if (error != 0)
return (error);
boot->br_rxlopri_lo =
htole32(TXP_ADDR_LO(sc->sc_ldata.txp_rxloring_paddr));
boot->br_rxlopri_hi =
htole32(TXP_ADDR_HI(sc->sc_ldata.txp_rxloring_paddr));
boot->br_rxlopri_siz =
htole32(RX_ENTRIES * sizeof(struct txp_rx_desc));
sc->sc_rxlor.r_tag = sc->sc_cdata.txp_rxloring_tag;
sc->sc_rxlor.r_map = sc->sc_cdata.txp_rxloring_map;
sc->sc_rxlor.r_desc = sc->sc_ldata.txp_rxloring;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sc->sc_rxlor.r_roff = &sc->sc_hostvar->hv_rx_lo_read_idx;
sc->sc_rxlor.r_woff = &sc->sc_hostvar->hv_rx_lo_write_idx;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Command ring. */
error = txp_dma_alloc(sc, "command ring",
&sc->sc_cdata.txp_cmdring_tag, sizeof(struct txp_cmd_desc), 0,
&sc->sc_cdata.txp_cmdring_map, (void **)&sc->sc_ldata.txp_cmdring,
sizeof(struct txp_cmd_desc) * CMD_ENTRIES,
&sc->sc_ldata.txp_cmdring_paddr);
if (error != 0)
return (error);
boot->br_cmd_lo = htole32(TXP_ADDR_LO(sc->sc_ldata.txp_cmdring_paddr));
boot->br_cmd_hi = htole32(TXP_ADDR_HI(sc->sc_ldata.txp_cmdring_paddr));
boot->br_cmd_siz = htole32(CMD_ENTRIES * sizeof(struct txp_cmd_desc));
sc->sc_cmdring.base = sc->sc_ldata.txp_cmdring;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sc->sc_cmdring.size = CMD_ENTRIES * sizeof(struct txp_cmd_desc);
sc->sc_cmdring.lastwrite = 0;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Response ring. */
error = txp_dma_alloc(sc, "response ring",
&sc->sc_cdata.txp_rspring_tag, sizeof(struct txp_rsp_desc), 0,
&sc->sc_cdata.txp_rspring_map, (void **)&sc->sc_ldata.txp_rspring,
sizeof(struct txp_rsp_desc) * RSP_ENTRIES,
&sc->sc_ldata.txp_rspring_paddr);
if (error != 0)
return (error);
boot->br_resp_lo = htole32(TXP_ADDR_LO(sc->sc_ldata.txp_rspring_paddr));
boot->br_resp_hi = htole32(TXP_ADDR_HI(sc->sc_ldata.txp_rspring_paddr));
boot->br_resp_siz = htole32(RSP_ENTRIES * sizeof(struct txp_rsp_desc));
sc->sc_rspring.base = sc->sc_ldata.txp_rspring;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sc->sc_rspring.size = RSP_ENTRIES * sizeof(struct txp_rsp_desc);
sc->sc_rspring.lastwrite = 0;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Receive buffer ring. */
error = txp_dma_alloc(sc, "receive buffer ring",
&sc->sc_cdata.txp_rxbufs_tag, sizeof(struct txp_rxbuf_desc), 0,
&sc->sc_cdata.txp_rxbufs_map, (void **)&sc->sc_ldata.txp_rxbufs,
sizeof(struct txp_rxbuf_desc) * RXBUF_ENTRIES,
&sc->sc_ldata.txp_rxbufs_paddr);
if (error != 0)
return (error);
boot->br_rxbuf_lo =
htole32(TXP_ADDR_LO(sc->sc_ldata.txp_rxbufs_paddr));
boot->br_rxbuf_hi =
htole32(TXP_ADDR_HI(sc->sc_ldata.txp_rxbufs_paddr));
boot->br_rxbuf_siz =
htole32(RXBUF_ENTRIES * sizeof(struct txp_rxbuf_desc));
sc->sc_rxbufs = sc->sc_ldata.txp_rxbufs;
/* Zero ring. */
error = txp_dma_alloc(sc, "zero buffer",
&sc->sc_cdata.txp_zero_tag, sizeof(uint32_t), 0,
&sc->sc_cdata.txp_zero_map, (void **)&sc->sc_ldata.txp_zero,
sizeof(uint32_t), &sc->sc_ldata.txp_zero_paddr);
if (error != 0)
return (error);
boot->br_zero_lo = htole32(TXP_ADDR_LO(sc->sc_ldata.txp_zero_paddr));
boot->br_zero_hi = htole32(TXP_ADDR_HI(sc->sc_ldata.txp_zero_paddr));
bus_dmamap_sync(sc->sc_cdata.txp_boot_tag, sc->sc_cdata.txp_boot_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
/* Create Tx buffers. */
error = bus_dma_tag_create(
sc->sc_cdata.txp_parent_tag, /* parent */
1, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
MCLBYTES * TXP_MAXTXSEGS, /* maxsize */
TXP_MAXTXSEGS, /* nsegments */
MCLBYTES, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->sc_cdata.txp_tx_tag);
if (error != 0) {
device_printf(sc->sc_dev, "could not create Tx DMA tag.\n");
goto fail;
}
/* Create tag for Rx buffers. */
error = bus_dma_tag_create(
sc->sc_cdata.txp_parent_tag, /* parent */
TXP_RXBUF_ALIGN, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
MCLBYTES, /* maxsize */
1, /* nsegments */
MCLBYTES, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->sc_cdata.txp_rx_tag);
if (error != 0) {
device_printf(sc->sc_dev, "could not create Rx DMA tag.\n");
goto fail;
}
/* Create DMA maps for Tx buffers. */
for (i = 0; i < TX_ENTRIES; i++) {
txd = &sc->sc_txd[i];
txd->sd_mbuf = NULL;
txd->sd_map = NULL;
error = bus_dmamap_create(sc->sc_cdata.txp_tx_tag, 0,
&txd->sd_map);
if (error != 0) {
device_printf(sc->sc_dev,
"could not create Tx dmamap.\n");
goto fail;
}
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Create DMA maps for Rx buffers. */
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
for (i = 0; i < RXBUF_ENTRIES; i++) {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sd = malloc(sizeof(struct txp_rx_swdesc), M_DEVBUF,
M_NOWAIT | M_ZERO);
if (sd == NULL) {
error = ENOMEM;
goto fail;
}
/*
* The virtual address part of descriptor is not used
* by hardware so use that to save an ring entry. We
* need bcopy here otherwise the address wouldn't be
* valid on big-endian architectures.
*/
rbd = sc->sc_rxbufs + i;
bcopy(&sd, (u_long *)&rbd->rb_vaddrlo, sizeof(sd));
Remember to zero out certain things that we malloc() and/or contigmalloc().
2001-07-27 19:38:56 +00:00
sd->sd_mbuf = NULL;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sd->sd_map = NULL;
error = bus_dmamap_create(sc->sc_cdata.txp_rx_tag, 0,
&sd->sd_map);
if (error != 0) {
device_printf(sc->sc_dev,
"could not create Rx dmamap.\n");
goto fail;
}
TAILQ_INSERT_TAIL(&sc->sc_free_list, sd, sd_next);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
fail:
return (error);
}
static void
txp_init_rings(struct txp_softc *sc)
{
bzero(sc->sc_ldata.txp_hostvar, sizeof(struct txp_hostvar));
bzero(sc->sc_ldata.txp_zero, sizeof(uint32_t));
sc->sc_txhir.r_cons = 0;
sc->sc_txhir.r_prod = 0;
sc->sc_txhir.r_cnt = 0;
sc->sc_txlor.r_cons = 0;
sc->sc_txlor.r_prod = 0;
sc->sc_txlor.r_cnt = 0;
sc->sc_cmdring.lastwrite = 0;
sc->sc_rspring.lastwrite = 0;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sc->sc_rxbufprod = 0;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sc->sc_cdata.txp_hostvar_tag,
sc->sc_cdata.txp_hostvar_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static int
txp_wait(struct txp_softc *sc, uint32_t state)
{
uint32_t reg;
int i;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
for (i = 0; i < TXP_TIMEOUT; i++) {
reg = READ_REG(sc, TXP_A2H_0);
if (reg == state)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
DELAY(50);
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
return (i == TXP_TIMEOUT ? ETIMEDOUT : 0);
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static void
txp_free_rings(struct txp_softc *sc)
{
struct txp_swdesc *txd;
struct txp_rx_swdesc *sd;
int i;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Tx buffers. */
if (sc->sc_cdata.txp_tx_tag != NULL) {
for (i = 0; i < TX_ENTRIES; i++) {
txd = &sc->sc_txd[i];
if (txd->sd_map != NULL) {
bus_dmamap_destroy(sc->sc_cdata.txp_tx_tag,
txd->sd_map);
txd->sd_map = NULL;
}
}
bus_dma_tag_destroy(sc->sc_cdata.txp_tx_tag);
sc->sc_cdata.txp_tx_tag = NULL;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Rx buffers. */
if (sc->sc_cdata.txp_rx_tag != NULL) {
if (sc->sc_rxbufs != NULL) {
KASSERT(TAILQ_FIRST(&sc->sc_busy_list) == NULL,
("%s : still have busy Rx buffers", __func__));
while ((sd = TAILQ_FIRST(&sc->sc_free_list)) != NULL) {
TAILQ_REMOVE(&sc->sc_free_list, sd, sd_next);
if (sd->sd_map != NULL) {
bus_dmamap_destroy(
sc->sc_cdata.txp_rx_tag,
sd->sd_map);
sd->sd_map = NULL;
}
free(sd, M_DEVBUF);
}
}
bus_dma_tag_destroy(sc->sc_cdata.txp_rx_tag);
sc->sc_cdata.txp_rx_tag = NULL;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Hi priority Tx ring. */
txp_dma_free(sc, &sc->sc_cdata.txp_txhiring_tag,
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
sc->sc_cdata.txp_txhiring_map,
(void **)&sc->sc_ldata.txp_txhiring,
&sc->sc_ldata.txp_txhiring_paddr);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Low priority Tx ring. */
txp_dma_free(sc, &sc->sc_cdata.txp_txloring_tag,
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
sc->sc_cdata.txp_txloring_map,
(void **)&sc->sc_ldata.txp_txloring,
&sc->sc_ldata.txp_txloring_paddr);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Hi priority Rx ring. */
txp_dma_free(sc, &sc->sc_cdata.txp_rxhiring_tag,
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
sc->sc_cdata.txp_rxhiring_map,
(void **)&sc->sc_ldata.txp_rxhiring,
&sc->sc_ldata.txp_rxhiring_paddr);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Low priority Rx ring. */
txp_dma_free(sc, &sc->sc_cdata.txp_rxloring_tag,
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
sc->sc_cdata.txp_rxloring_map,
(void **)&sc->sc_ldata.txp_rxloring,
&sc->sc_ldata.txp_rxloring_paddr);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Receive buffer ring. */
txp_dma_free(sc, &sc->sc_cdata.txp_rxbufs_tag,
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
sc->sc_cdata.txp_rxbufs_map, (void **)&sc->sc_ldata.txp_rxbufs,
&sc->sc_ldata.txp_rxbufs_paddr);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Command ring. */
txp_dma_free(sc, &sc->sc_cdata.txp_cmdring_tag,
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
sc->sc_cdata.txp_cmdring_map, (void **)&sc->sc_ldata.txp_cmdring,
&sc->sc_ldata.txp_cmdring_paddr);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Response ring. */
txp_dma_free(sc, &sc->sc_cdata.txp_rspring_tag,
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
sc->sc_cdata.txp_rspring_map, (void **)&sc->sc_ldata.txp_rspring,
&sc->sc_ldata.txp_rspring_paddr);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Zero ring. */
txp_dma_free(sc, &sc->sc_cdata.txp_zero_tag,
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
sc->sc_cdata.txp_zero_map, (void **)&sc->sc_ldata.txp_zero,
&sc->sc_ldata.txp_zero_paddr);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Host variables. */
txp_dma_free(sc, &sc->sc_cdata.txp_hostvar_tag,
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
sc->sc_cdata.txp_hostvar_map, (void **)&sc->sc_ldata.txp_hostvar,
&sc->sc_ldata.txp_hostvar_paddr);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Boot record. */
txp_dma_free(sc, &sc->sc_cdata.txp_boot_tag,
Fix teardown of static DMA allocations in various NIC drivers: - Add missing calls to bus_dmamap_unload() in et(4). - Check the bus address against 0 to decide when to call bus_dmamap_unload() instead of comparing the bus_dma map against NULL. - Check the virtual address against NULL to decide when to call bus_dmamem_free() instead of comparing the bus_dma map against NULL. - Don't clear bus_dma map pointers to NULL for static allocations. Instead, treat the value as completely opaque. - Pass the correct virtual address to bus_dmamem_free() in wpi(4) instead of trying to free a pointer to the virtual address. Reviewed by: yongari
2014-06-17 14:47:49 +00:00
sc->sc_cdata.txp_boot_map, (void **)&sc->sc_ldata.txp_boot,
&sc->sc_ldata.txp_boot_paddr);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (sc->sc_cdata.txp_parent_tag != NULL) {
bus_dma_tag_destroy(sc->sc_cdata.txp_parent_tag);
sc->sc_cdata.txp_parent_tag = NULL;
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
struct txp_softc *sc = ifp->if_softc;
struct ifreq *ifr = (struct ifreq *)data;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
int capenable, error = 0, mask;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
switch(command) {
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
case SIOCSIFFLAGS:
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if ((ifp->if_flags & IFF_UP) != 0) {
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
if (((ifp->if_flags ^ sc->sc_if_flags)
& (IFF_PROMISC | IFF_ALLMULTI)) != 0)
txp_set_filter(sc);
} else {
if ((sc->sc_flags & TXP_FLAG_DETACH) == 0)
txp_init_locked(sc);
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
} else {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
txp_stop(sc);
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sc->sc_if_flags = ifp->if_flags;
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_UNLOCK(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
/*
* Multicast list has changed; set the hardware
* filter accordingly.
*/
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
txp_set_filter(sc);
TXP_UNLOCK(sc);
break;
case SIOCSIFCAP:
TXP_LOCK(sc);
capenable = ifp->if_capenable;
mask = ifr->ifr_reqcap ^ ifp->if_capenable;
if ((mask & IFCAP_TXCSUM) != 0 &&
(ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
ifp->if_capenable ^= IFCAP_TXCSUM;
if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
ifp->if_hwassist |= TXP_CSUM_FEATURES;
else
ifp->if_hwassist &= ~TXP_CSUM_FEATURES;
}
if ((mask & IFCAP_RXCSUM) != 0 &&
(ifp->if_capabilities & IFCAP_RXCSUM) != 0)
ifp->if_capenable ^= IFCAP_RXCSUM;
if ((mask & IFCAP_WOL_MAGIC) != 0 &&
(ifp->if_capabilities & IFCAP_WOL_MAGIC) != 0)
ifp->if_capenable ^= IFCAP_WOL_MAGIC;
if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
(ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0)
ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
if ((mask & IFCAP_VLAN_HWCSUM) != 0 &&
(ifp->if_capabilities & IFCAP_VLAN_HWCSUM) != 0)
ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
if ((ifp->if_capenable & IFCAP_TXCSUM) == 0)
ifp->if_capenable &= ~IFCAP_VLAN_HWCSUM;
if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
ifp->if_capenable &= ~IFCAP_VLAN_HWCSUM;
if (capenable != ifp->if_capenable)
txp_set_capabilities(sc);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_UNLOCK(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
VLAN_CAPABILITIES(ifp);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
case SIOCGIFMEDIA:
case SIOCSIFMEDIA:
error = ifmedia_ioctl(ifp, ifr, &sc->sc_ifmedia, command);
break;
default:
network interface driver changes: o don't strip the Ethernet header from inbound packets; pass packets up the stack intact (required significant changes to some drivers) o reference common definitions in net/ethernet.h (e.g. ETHER_ALIGN) o track ether_ifattach/ether_ifdetach API changes o track bpf changes (use BPF_TAP and BPF_MTAP) o track vlan changes (ifnet capabilities, revised processing scheme, etc.) o use if_input to pass packets "up" o call ether_ioctl for default handling of ioctls Reviewed by: many Approved by: re
2002-11-14 23:54:55 +00:00
error = ether_ioctl(ifp, command, data);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
}
style(9) - space after keywords.
2009-03-11 08:49:17 +00:00
return (error);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_rxring_fill(struct txp_softc *sc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_rxbuf_desc *rbd;
struct txp_rx_swdesc *sd;
bus_dma_segment_t segs[1];
int error, i, nsegs;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK_ASSERT(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sc->sc_cdata.txp_rxbufs_tag,
sc->sc_cdata.txp_rxbufs_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
for (i = 0; i < RXBUF_ENTRIES; i++) {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sd = TAILQ_FIRST(&sc->sc_free_list);
if (sd == NULL)
return (ENOMEM);
rbd = sc->sc_rxbufs + i;
bcopy(&sd, (u_long *)&rbd->rb_vaddrlo, sizeof(sd));
KASSERT(sd->sd_mbuf == NULL,
("%s : Rx buffer ring corrupted", __func__));
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
sd->sd_mbuf = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (sd->sd_mbuf == NULL)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
return (ENOMEM);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sd->sd_mbuf->m_pkthdr.len = sd->sd_mbuf->m_len = MCLBYTES;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#ifndef __NO_STRICT_ALIGNMENT
m_adj(sd->sd_mbuf, TXP_RXBUF_ALIGN);
#endif
if ((error = bus_dmamap_load_mbuf_sg(sc->sc_cdata.txp_rx_tag,
sd->sd_map, sd->sd_mbuf, segs, &nsegs, 0)) != 0) {
m_freem(sd->sd_mbuf);
sd->sd_mbuf = NULL;
return (error);
}
KASSERT(nsegs == 1, ("%s : %d segments returned!", __func__,
nsegs));
TAILQ_REMOVE(&sc->sc_free_list, sd, sd_next);
TAILQ_INSERT_TAIL(&sc->sc_busy_list, sd, sd_next);
bus_dmamap_sync(sc->sc_cdata.txp_rx_tag, sd->sd_map,
BUS_DMASYNC_PREREAD);
rbd->rb_paddrlo = htole32(TXP_ADDR_LO(segs[0].ds_addr));
rbd->rb_paddrhi = htole32(TXP_ADDR_HI(segs[0].ds_addr));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sc->sc_cdata.txp_rxbufs_tag,
sc->sc_cdata.txp_rxbufs_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
sc->sc_rxbufprod = RXBUF_ENTRIES - 1;
sc->sc_hostvar->hv_rx_buf_write_idx =
htole32(TXP_IDX2OFFSET(RXBUF_ENTRIES - 1));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
style(9) - space after keywords.
2009-03-11 08:49:17 +00:00
return (0);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_rxring_empty(struct txp_softc *sc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_rx_swdesc *sd;
int cnt;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK_ASSERT(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (sc->sc_rxbufs == NULL)
return;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sc->sc_cdata.txp_hostvar_tag,
sc->sc_cdata.txp_hostvar_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
/* Release allocated Rx buffers. */
cnt = 0;
while ((sd = TAILQ_FIRST(&sc->sc_busy_list)) != NULL) {
TAILQ_REMOVE(&sc->sc_busy_list, sd, sd_next);
KASSERT(sd->sd_mbuf != NULL,
("%s : Rx buffer ring corrupted", __func__));
bus_dmamap_sync(sc->sc_cdata.txp_rx_tag, sd->sd_map,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->sc_cdata.txp_rx_tag, sd->sd_map);
m_freem(sd->sd_mbuf);
sd->sd_mbuf = NULL;
TAILQ_INSERT_TAIL(&sc->sc_free_list, sd, sd_next);
cnt++;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
}
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_init(void *xsc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
struct txp_softc *sc;
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
sc = xsc;
TXP_LOCK(sc);
txp_init_locked(sc);
TXP_UNLOCK(sc);
}
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_init_locked(struct txp_softc *sc)
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
{
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
struct ifnet *ifp;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
uint8_t *eaddr;
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint16_t p1;
uint32_t p2;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
int error;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK_ASSERT(sc);
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ifp = sc->sc_ifp;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Initialize ring structure. */
txp_init_rings(sc);
/* Wakeup controller. */
WRITE_REG(sc, TXP_H2A_0, TXP_BOOTCMD_WAKEUP);
TXP_BARRIER(sc, TXP_H2A_0, 4, BUS_SPACE_BARRIER_WRITE);
/*
* It seems that earlier NV image can go back to online from
* wakeup command but newer ones require controller reset.
* So jut reset controller again.
*/
if (txp_reset(sc) != 0)
goto init_fail;
/* Download firmware. */
error = txp_download_fw(sc);
if (error != 0) {
device_printf(sc->sc_dev, "could not download firmware.\n");
goto init_fail;
}
bus_dmamap_sync(sc->sc_cdata.txp_hostvar_tag,
sc->sc_cdata.txp_hostvar_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
if ((error = txp_rxring_fill(sc)) != 0) {
device_printf(sc->sc_dev, "no memory for Rx buffers.\n");
goto init_fail;
}
bus_dmamap_sync(sc->sc_cdata.txp_hostvar_tag,
sc->sc_cdata.txp_hostvar_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
if (txp_boot(sc, STAT_WAITING_FOR_BOOT) != 0) {
device_printf(sc->sc_dev, "could not boot firmware.\n");
goto init_fail;
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/*
* Quite contrary to Typhoon T2 software functional specification,
* it seems that TXP_CMD_RECV_BUFFER_CONTROL command is not
* implemented in the firmware. This means driver should have to
* handle misaligned frames on alignment architectures. AFAIK this
* is the only controller manufactured by 3Com that has this stupid
* bug. 3Com should fix this.
*/
if (txp_command(sc, TXP_CMD_MAX_PKT_SIZE_WRITE, TXP_MAX_PKTLEN, 0, 0,
NULL, NULL, NULL, TXP_CMD_NOWAIT) != 0)
goto init_fail;
/* Undocumented command(interrupt coalescing disable?) - From Linux. */
if (txp_command(sc, TXP_CMD_FILTER_DEFINE, 0, 0, 0, NULL, NULL, NULL,
TXP_CMD_NOWAIT) != 0)
goto init_fail;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
/* Set station address. */
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
eaddr = IF_LLADDR(sc->sc_ifp);
p1 = 0;
((uint8_t *)&p1)[1] = eaddr[0];
((uint8_t *)&p1)[0] = eaddr[1];
p1 = le16toh(p1);
((uint8_t *)&p2)[3] = eaddr[2];
((uint8_t *)&p2)[2] = eaddr[3];
((uint8_t *)&p2)[1] = eaddr[4];
((uint8_t *)&p2)[0] = eaddr[5];
p2 = le32toh(p2);
if (txp_command(sc, TXP_CMD_STATION_ADDRESS_WRITE, p1, p2, 0,
NULL, NULL, NULL, TXP_CMD_NOWAIT) != 0)
goto init_fail;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
txp_set_filter(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_set_capabilities(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (txp_command(sc, TXP_CMD_CLEAR_STATISTICS, 0, 0, 0,
NULL, NULL, NULL, TXP_CMD_NOWAIT))
goto init_fail;
if (txp_command(sc, TXP_CMD_XCVR_SELECT, sc->sc_xcvr, 0, 0,
NULL, NULL, NULL, TXP_CMD_NOWAIT) != 0)
goto init_fail;
if (txp_command(sc, TXP_CMD_TX_ENABLE, 0, 0, 0, NULL, NULL, NULL,
TXP_CMD_NOWAIT) != 0)
goto init_fail;
if (txp_command(sc, TXP_CMD_RX_ENABLE, 0, 0, 0, NULL, NULL, NULL,
TXP_CMD_NOWAIT) != 0)
goto init_fail;
/* Ack all pending interrupts and enable interrupts. */
WRITE_REG(sc, TXP_ISR, TXP_INTR_ALL);
WRITE_REG(sc, TXP_IER, TXP_INTRS);
WRITE_REG(sc, TXP_IMR, TXP_INTR_NONE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
ifp->if_drv_flags |= IFF_DRV_RUNNING;
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
callout_reset(&sc->sc_tick, hz, txp_tick, sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
return;
init_fail:
txp_rxring_empty(sc);
txp_init_rings(sc);
txp_reset(sc);
WRITE_REG(sc, TXP_IMR, TXP_INTR_ALL);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_tick(void *vsc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_softc *sc;
struct ifnet *ifp;
struct txp_rsp_desc *rsp;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
struct txp_ext_desc *ext;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
int link;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sc = vsc;
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK_ASSERT(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sc->sc_cdata.txp_hostvar_tag,
sc->sc_cdata.txp_hostvar_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
txp_rxbuf_reclaim(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sc->sc_cdata.txp_hostvar_tag,
sc->sc_cdata.txp_hostvar_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
ifp = sc->sc_ifp;
rsp = NULL;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
link = sc->sc_flags & TXP_FLAG_LINK;
if (txp_ext_command(sc, TXP_CMD_READ_STATISTICS, 0, 0, 0, NULL, 0,
&rsp, TXP_CMD_WAIT))
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
goto out;
if (rsp->rsp_numdesc != 6)
goto out;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_stats_update(sc, rsp);
if (link == 0 && (sc->sc_flags & TXP_FLAG_LINK) != 0) {
ext = (struct txp_ext_desc *)(rsp + 1);
/* Update baudrate with resolved speed. */
if ((ext[5].ext_2 & 0x02) != 0)
ifp->if_baudrate = IF_Mbps(100);
else
ifp->if_baudrate = IF_Mbps(10);
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
out:
if (rsp != NULL)
free(rsp, M_DEVBUF);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_watchdog(sc);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
callout_reset(&sc->sc_tick, hz, txp_tick, sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_start(struct ifnet *ifp)
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
{
struct txp_softc *sc;
sc = ifp->if_softc;
TXP_LOCK(sc);
txp_start_locked(ifp);
TXP_UNLOCK(sc);
}
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_start_locked(struct ifnet *ifp)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_softc *sc;
struct mbuf *m_head;
int enq;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sc = ifp->if_softc;
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK_ASSERT(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
IFF_DRV_RUNNING || (sc->sc_flags & TXP_FLAG_LINK) == 0)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
for (enq = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd); ) {
IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
if (m_head == NULL)
break;
/*
* Pack the data into the transmit ring. If we
* don't have room, set the OACTIVE flag and wait
* for the NIC to drain the ring.
* ATM only Hi-ring is used.
*/
if (txp_encap(sc, &sc->sc_txhir, &m_head)) {
if (m_head == NULL)
break;
IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/*
* If there's a BPF listener, bounce a copy of this frame
* to him.
*/
ETHER_BPF_MTAP(ifp, m_head);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Send queued frame. */
WRITE_REG(sc, sc->sc_txhir.r_reg,
TXP_IDX2OFFSET(sc->sc_txhir.r_prod));
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (enq > 0) {
/* Set a timeout in case the chip goes out to lunch. */
sc->sc_watchdog_timer = TXP_TX_TIMEOUT;
}
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static int
txp_encap(struct txp_softc *sc, struct txp_tx_ring *r, struct mbuf **m_head)
{
struct txp_tx_desc *first_txd;
struct txp_frag_desc *fxd;
struct txp_swdesc *sd;
struct mbuf *m;
bus_dma_segment_t txsegs[TXP_MAXTXSEGS];
int error, i, nsegs;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_LOCK_ASSERT(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
M_ASSERTPKTHDR((*m_head));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
m = *m_head;
first_txd = r->r_desc + r->r_prod;
sd = sc->sc_txd + r->r_prod;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
error = bus_dmamap_load_mbuf_sg(sc->sc_cdata.txp_tx_tag, sd->sd_map,
*m_head, txsegs, &nsegs, 0);
if (error == EFBIG) {
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m = m_collapse(*m_head, M_NOWAIT, TXP_MAXTXSEGS);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (m == NULL) {
m_freem(*m_head);
*m_head = NULL;
return (ENOMEM);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
*m_head = m;
error = bus_dmamap_load_mbuf_sg(sc->sc_cdata.txp_tx_tag,
sd->sd_map, *m_head, txsegs, &nsegs, 0);
if (error != 0) {
m_freem(*m_head);
*m_head = NULL;
return (error);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
} else if (error != 0)
return (error);
if (nsegs == 0) {
m_freem(*m_head);
*m_head = NULL;
return (EIO);
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Check descriptor overrun. */
if (r->r_cnt + nsegs >= TX_ENTRIES - TXP_TXD_RESERVED) {
bus_dmamap_unload(sc->sc_cdata.txp_tx_tag, sd->sd_map);
return (ENOBUFS);
}
bus_dmamap_sync(sc->sc_cdata.txp_tx_tag, sd->sd_map,
BUS_DMASYNC_PREWRITE);
sd->sd_mbuf = m;
first_txd->tx_flags = TX_FLAGS_TYPE_DATA;
first_txd->tx_numdesc = 0;
first_txd->tx_addrlo = 0;
first_txd->tx_addrhi = 0;
first_txd->tx_totlen = 0;
first_txd->tx_pflags = 0;
r->r_cnt++;
TXP_DESC_INC(r->r_prod, TX_ENTRIES);
/* Configure Tx IP/TCP/UDP checksum offload. */
if ((m->m_pkthdr.csum_flags & CSUM_IP) != 0)
first_txd->tx_pflags |= htole32(TX_PFLAGS_IPCKSUM);
#ifdef notyet
/* XXX firmware bug. */
if ((m->m_pkthdr.csum_flags & CSUM_TCP) != 0)
first_txd->tx_pflags |= htole32(TX_PFLAGS_TCPCKSUM);
if ((m->m_pkthdr.csum_flags & CSUM_UDP) != 0)
first_txd->tx_pflags |= htole32(TX_PFLAGS_UDPCKSUM);
#endif
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Configure VLAN hardware tag insertion. */
if ((m->m_flags & M_VLANTAG) != 0)
first_txd->tx_pflags |=
htole32(TX_PFLAGS_VLAN | TX_PFLAGS_PRIO |
(bswap16(m->m_pkthdr.ether_vtag) << TX_PFLAGS_VLANTAG_S));
for (i = 0; i < nsegs; i++) {
fxd = (struct txp_frag_desc *)(r->r_desc + r->r_prod);
fxd->frag_flags = FRAG_FLAGS_TYPE_FRAG | TX_FLAGS_VALID;
fxd->frag_rsvd1 = 0;
fxd->frag_len = htole16(txsegs[i].ds_len);
fxd->frag_addrhi = htole32(TXP_ADDR_HI(txsegs[i].ds_addr));
fxd->frag_addrlo = htole32(TXP_ADDR_LO(txsegs[i].ds_addr));
fxd->frag_rsvd2 = 0;
first_txd->tx_numdesc++;
r->r_cnt++;
TXP_DESC_INC(r->r_prod, TX_ENTRIES);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Lastly set valid flag. */
first_txd->tx_flags |= TX_FLAGS_VALID;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Sync descriptors. */
bus_dmamap_sync(r->r_tag, r->r_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
return (0);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
/*
* Handle simple commands sent to the typhoon
*/
static int
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
txp_command(struct txp_softc *sc, uint16_t id, uint16_t in1, uint32_t in2,
uint32_t in3, uint16_t *out1, uint32_t *out2, uint32_t *out3, int wait)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_rsp_desc *rsp;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
rsp = NULL;
if (txp_ext_command(sc, id, in1, in2, in3, NULL, 0, &rsp, wait) != 0) {
device_printf(sc->sc_dev, "command 0x%02x failed\n", id);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return (-1);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (wait == TXP_CMD_NOWAIT)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return (0);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
KASSERT(rsp != NULL, ("rsp is NULL!\n"));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (out1 != NULL)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
*out1 = le16toh(rsp->rsp_par1);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (out2 != NULL)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
*out2 = le32toh(rsp->rsp_par2);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (out3 != NULL)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
*out3 = le32toh(rsp->rsp_par3);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
free(rsp, M_DEVBUF);
return (0);
}
static int
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_ext_command(struct txp_softc *sc, uint16_t id, uint16_t in1, uint32_t in2,
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint32_t in3, struct txp_ext_desc *in_extp, uint8_t in_extn,
Change the functions to ANSI in those cases where it breaks promotion to int rule. See ISO C Standard: SS6.7.5.3:15. Approved by: kib (mentor) Reviewed by: warner Tested by: silence on -current
2009-02-24 18:09:31 +00:00
struct txp_rsp_desc **rspp, int wait)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_hostvar *hv;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
struct txp_cmd_desc *cmd;
struct txp_ext_desc *ext;
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint32_t idx, i;
uint16_t seq;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
int error;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
error = 0;
hv = sc->sc_hostvar;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (txp_cmd_desc_numfree(sc) < (in_extn + 1)) {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
device_printf(sc->sc_dev,
"%s : out of free cmd descriptors for command 0x%02x\n",
__func__, id);
return (ENOBUFS);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sc->sc_cdata.txp_cmdring_tag,
sc->sc_cdata.txp_cmdring_map, BUS_DMASYNC_POSTWRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
idx = sc->sc_cmdring.lastwrite;
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
cmd = (struct txp_cmd_desc *)(((uint8_t *)sc->sc_cmdring.base) + idx);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bzero(cmd, sizeof(*cmd));
cmd->cmd_numdesc = in_extn;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
seq = sc->sc_seq++;
cmd->cmd_seq = htole16(seq);
cmd->cmd_id = htole16(id);
cmd->cmd_par1 = htole16(in1);
cmd->cmd_par2 = htole32(in2);
cmd->cmd_par3 = htole32(in3);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
cmd->cmd_flags = CMD_FLAGS_TYPE_CMD |
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
(wait == TXP_CMD_WAIT ? CMD_FLAGS_RESP : 0) | CMD_FLAGS_VALID;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
idx += sizeof(struct txp_cmd_desc);
if (idx == sc->sc_cmdring.size)
idx = 0;
for (i = 0; i < in_extn; i++) {
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
ext = (struct txp_ext_desc *)(((uint8_t *)sc->sc_cmdring.base) + idx);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bcopy(in_extp, ext, sizeof(struct txp_ext_desc));
in_extp++;
idx += sizeof(struct txp_cmd_desc);
if (idx == sc->sc_cmdring.size)
idx = 0;
}
sc->sc_cmdring.lastwrite = idx;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sc->sc_cdata.txp_cmdring_tag,
sc->sc_cdata.txp_cmdring_map, BUS_DMASYNC_PREWRITE);
bus_dmamap_sync(sc->sc_cdata.txp_hostvar_tag,
sc->sc_cdata.txp_hostvar_map, BUS_DMASYNC_PREREAD |
BUS_DMASYNC_PREWRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
WRITE_REG(sc, TXP_H2A_2, sc->sc_cmdring.lastwrite);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_BARRIER(sc, TXP_H2A_2, 4, BUS_SPACE_BARRIER_WRITE);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (wait == TXP_CMD_NOWAIT)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return (0);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
for (i = 0; i < TXP_TIMEOUT; i++) {
bus_dmamap_sync(sc->sc_cdata.txp_hostvar_tag,
sc->sc_cdata.txp_hostvar_map, BUS_DMASYNC_POSTREAD |
BUS_DMASYNC_POSTWRITE);
if (le32toh(hv->hv_resp_read_idx) !=
le32toh(hv->hv_resp_write_idx)) {
error = txp_response(sc, id, seq, rspp);
bus_dmamap_sync(sc->sc_cdata.txp_hostvar_tag,
sc->sc_cdata.txp_hostvar_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
if (error != 0)
return (error);
if (*rspp != NULL)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
}
DELAY(50);
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (i == TXP_TIMEOUT) {
device_printf(sc->sc_dev, "command 0x%02x timedout\n", id);
error = ETIMEDOUT;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
return (error);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static int
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_response(struct txp_softc *sc, uint16_t id, uint16_t seq,
Change the functions to ANSI in those cases where it breaks promotion to int rule. See ISO C Standard: SS6.7.5.3:15. Approved by: kib (mentor) Reviewed by: warner Tested by: silence on -current
2009-02-24 18:09:31 +00:00
struct txp_rsp_desc **rspp)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_hostvar *hv;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
struct txp_rsp_desc *rsp;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
uint32_t ridx;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sc->sc_cdata.txp_rspring_tag,
sc->sc_cdata.txp_rspring_map, BUS_DMASYNC_POSTREAD);
hv = sc->sc_hostvar;
ridx = le32toh(hv->hv_resp_read_idx);
while (ridx != le32toh(hv->hv_resp_write_idx)) {
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
rsp = (struct txp_rsp_desc *)(((uint8_t *)sc->sc_rspring.base) + ridx);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (id == le16toh(rsp->rsp_id) &&
le16toh(rsp->rsp_seq) == seq) {
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
*rspp = (struct txp_rsp_desc *)malloc(
sizeof(struct txp_rsp_desc) * (rsp->rsp_numdesc + 1),
M_DEVBUF, M_NOWAIT);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (*rspp == NULL) {
device_printf(sc->sc_dev,"%s : command 0x%02x "
"memory allocation failure\n",
__func__, id);
return (ENOMEM);
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
txp_rsp_fixup(sc, rsp, *rspp);
return (0);
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if ((rsp->rsp_flags & RSP_FLAGS_ERROR) != 0) {
device_printf(sc->sc_dev,
"%s : command 0x%02x response error!\n", __func__,
le16toh(rsp->rsp_id));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
txp_rsp_fixup(sc, rsp, NULL);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ridx = le32toh(hv->hv_resp_read_idx);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
continue;
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/*
* The following unsolicited responses are handled during
* processing of TXP_CMD_READ_STATISTICS which requires
* response. Driver abuses the command to detect media
* status change.
* TXP_CMD_FILTER_DEFINE is not an unsolicited response
* but we don't process response ring in interrupt handler
* so we have to ignore this command here, otherwise
* unknown command message would be printed.
*/
switch (le16toh(rsp->rsp_id)) {
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
case TXP_CMD_CYCLE_STATISTICS:
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
case TXP_CMD_FILTER_DEFINE:
break;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
case TXP_CMD_MEDIA_STATUS_READ:
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if ((le16toh(rsp->rsp_par1) & 0x0800) == 0) {
sc->sc_flags |= TXP_FLAG_LINK;
if_link_state_change(sc->sc_ifp,
LINK_STATE_UP);
} else {
sc->sc_flags &= ~TXP_FLAG_LINK;
if_link_state_change(sc->sc_ifp,
LINK_STATE_DOWN);
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
case TXP_CMD_HELLO_RESPONSE:
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/*
* Driver should repsond to hello message but
* TXP_CMD_READ_STATISTICS is issued for every
* hz, therefore there is no need to send an
* explicit command here.
*/
device_printf(sc->sc_dev, "%s : hello\n", __func__);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
default:
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
device_printf(sc->sc_dev,
"%s : unknown command 0x%02x\n", __func__,
le16toh(rsp->rsp_id));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
txp_rsp_fixup(sc, rsp, NULL);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ridx = le32toh(hv->hv_resp_read_idx);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
return (0);
}
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_rsp_fixup(struct txp_softc *sc, struct txp_rsp_desc *rsp,
struct txp_rsp_desc *dst)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_rsp_desc *src;
struct txp_hostvar *hv;
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint32_t i, ridx;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
src = rsp;
hv = sc->sc_hostvar;
ridx = le32toh(hv->hv_resp_read_idx);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
for (i = 0; i < rsp->rsp_numdesc + 1; i++) {
if (dst != NULL)
bcopy(src, dst++, sizeof(struct txp_rsp_desc));
ridx += sizeof(struct txp_rsp_desc);
if (ridx == sc->sc_rspring.size) {
src = sc->sc_rspring.base;
ridx = 0;
} else
src++;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sc->sc_rspring.lastwrite = ridx;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
Remove extra tab characters.
2009-03-11 08:28:24 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
hv->hv_resp_read_idx = htole32(ridx);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_cmd_desc_numfree(struct txp_softc *sc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_hostvar *hv;
struct txp_boot_record *br;
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint32_t widx, ridx, nfree;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
bus_dmamap_sync(sc->sc_cdata.txp_hostvar_tag,
sc->sc_cdata.txp_hostvar_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
hv = sc->sc_hostvar;
br = sc->sc_boot;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
widx = sc->sc_cmdring.lastwrite;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ridx = le32toh(hv->hv_cmd_read_idx);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (widx == ridx) {
/* Ring is completely free */
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
nfree = le32toh(br->br_cmd_siz) - sizeof(struct txp_cmd_desc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
} else {
if (widx > ridx)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
nfree = le32toh(br->br_cmd_siz) -
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
(widx - ridx + sizeof(struct txp_cmd_desc));
else
nfree = ridx - widx - sizeof(struct txp_cmd_desc);
}
return (nfree / sizeof(struct txp_cmd_desc));
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static int
txp_sleep(struct txp_softc *sc, int capenable)
{
uint16_t events;
int error;
events = 0;
if ((capenable & IFCAP_WOL_MAGIC) != 0)
events |= 0x01;
error = txp_command(sc, TXP_CMD_ENABLE_WAKEUP_EVENTS, events, 0, 0,
NULL, NULL, NULL, TXP_CMD_NOWAIT);
if (error == 0) {
/* Goto sleep. */
error = txp_command(sc, TXP_CMD_GOTO_SLEEP, 0, 0, 0, NULL,
NULL, NULL, TXP_CMD_NOWAIT);
if (error == 0) {
error = txp_wait(sc, STAT_SLEEPING);
if (error != 0)
device_printf(sc->sc_dev,
"unable to enter into sleep\n");
}
}
return (error);
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_stop(struct txp_softc *sc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
struct ifnet *ifp;
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK_ASSERT(sc);
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ifp = sc->sc_ifp;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
return;
WRITE_REG(sc, TXP_IER, TXP_INTR_NONE);
WRITE_REG(sc, TXP_ISR, TXP_INTR_ALL);
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
sc->sc_flags &= ~TXP_FLAG_LINK;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
callout_stop(&sc->sc_tick);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_command(sc, TXP_CMD_TX_DISABLE, 0, 0, 0, NULL, NULL, NULL,
TXP_CMD_NOWAIT);
txp_command(sc, TXP_CMD_RX_DISABLE, 0, 0, 0, NULL, NULL, NULL,
TXP_CMD_NOWAIT);
/* Save statistics for later use. */
txp_stats_save(sc);
/* Halt controller. */
txp_command(sc, TXP_CMD_HALT, 0, 0, 0, NULL, NULL, NULL,
TXP_CMD_NOWAIT);
if (txp_wait(sc, STAT_HALTED) != 0)
device_printf(sc->sc_dev, "controller halt timedout!\n");
/* Reclaim Tx/Rx buffers. */
if (sc->sc_txhir.r_cnt && (sc->sc_txhir.r_cons !=
TXP_OFFSET2IDX(le32toh(*(sc->sc_txhir.r_off)))))
txp_tx_reclaim(sc, &sc->sc_txhir);
if (sc->sc_txlor.r_cnt && (sc->sc_txlor.r_cons !=
TXP_OFFSET2IDX(le32toh(*(sc->sc_txlor.r_off)))))
txp_tx_reclaim(sc, &sc->sc_txlor);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
txp_rxring_empty(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_init_rings(sc);
/* Reset controller and make it reload sleep image. */
txp_reset(sc);
/* Let controller boot from sleep image. */
if (txp_boot(sc, STAT_WAITING_FOR_HOST_REQUEST) != 0)
device_printf(sc->sc_dev, "could not boot sleep image\n");
txp_sleep(sc, 0);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static void
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_watchdog(struct txp_softc *sc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct ifnet *ifp;
TXP_LOCK_ASSERT(sc);
Remove return statement at the end of function that returns void.
2009-03-11 08:41:57 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (sc->sc_watchdog_timer == 0 || --sc->sc_watchdog_timer)
return;
ifp = sc->sc_ifp;
if_printf(ifp, "watchdog timeout -- resetting\n");
- Provide txp_get_counter() to return counters that are not collected, but taken from hardware. - Mechanically convert to if_inc_counter() the rest of counters.
2014-09-24 11:58:23 +00:00
if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_stop(sc);
txp_init_locked(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static int
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_ifmedia_upd(struct ifnet *ifp)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
struct txp_softc *sc = ifp->if_softc;
struct ifmedia *ifm = &sc->sc_ifmedia;
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint16_t new_xcvr;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK(sc);
if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) {
TXP_UNLOCK(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return (EINVAL);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (IFM_SUBTYPE(ifm->ifm_media) == IFM_10_T) {
if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
new_xcvr = TXP_XCVR_10_FDX;
else
new_xcvr = TXP_XCVR_10_HDX;
} else if (IFM_SUBTYPE(ifm->ifm_media) == IFM_100_TX) {
if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
new_xcvr = TXP_XCVR_100_FDX;
else
new_xcvr = TXP_XCVR_100_HDX;
} else if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
new_xcvr = TXP_XCVR_AUTO;
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
} else {
TXP_UNLOCK(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return (EINVAL);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
/* nothing to do */
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
if (sc->sc_xcvr == new_xcvr) {
TXP_UNLOCK(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return (0);
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
txp_command(sc, TXP_CMD_XCVR_SELECT, new_xcvr, 0, 0,
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
NULL, NULL, NULL, TXP_CMD_NOWAIT);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
sc->sc_xcvr = new_xcvr;
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_UNLOCK(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
return (0);
}
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
struct txp_softc *sc = ifp->if_softc;
struct ifmedia *ifm = &sc->sc_ifmedia;
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint16_t bmsr, bmcr, anar, anlpar;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
ifmr->ifm_status = IFM_AVALID;
ifmr->ifm_active = IFM_ETHER;
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_LOCK(sc);
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/* Check whether firmware is running. */
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
goto bail;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (txp_command(sc, TXP_CMD_PHY_MGMT_READ, 0, MII_BMSR, 0,
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
&bmsr, NULL, NULL, TXP_CMD_WAIT))
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
goto bail;
if (txp_command(sc, TXP_CMD_PHY_MGMT_READ, 0, MII_BMSR, 0,
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
&bmsr, NULL, NULL, TXP_CMD_WAIT))
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
goto bail;
if (txp_command(sc, TXP_CMD_PHY_MGMT_READ, 0, MII_BMCR, 0,
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
&bmcr, NULL, NULL, TXP_CMD_WAIT))
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
goto bail;
if (txp_command(sc, TXP_CMD_PHY_MGMT_READ, 0, MII_ANLPAR, 0,
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
&anlpar, NULL, NULL, TXP_CMD_WAIT))
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
goto bail;
Read MII_ANAR register and get common denominator ability. PR: 92599
2007-11-16 10:32:10 +00:00
if (txp_command(sc, TXP_CMD_PHY_MGMT_READ, 0, MII_ANAR, 0,
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
&anar, NULL, NULL, TXP_CMD_WAIT))
Read MII_ANAR register and get common denominator ability. PR: 92599
2007-11-16 10:32:10 +00:00
goto bail;
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_UNLOCK(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
if (bmsr & BMSR_LINK)
ifmr->ifm_status |= IFM_ACTIVE;
if (bmcr & BMCR_ISO) {
ifmr->ifm_active |= IFM_NONE;
ifmr->ifm_status = 0;
return;
}
if (bmcr & BMCR_LOOP)
ifmr->ifm_active |= IFM_LOOP;
if (bmcr & BMCR_AUTOEN) {
if ((bmsr & BMSR_ACOMP) == 0) {
ifmr->ifm_active |= IFM_NONE;
return;
}
Read MII_ANAR register and get common denominator ability. PR: 92599
2007-11-16 10:32:10 +00:00
anlpar &= anar;
IEEE 802.3 Annex 28B.3 explicitly specifies the following relative priorities of the technologies supported by 802.3 Selector Field value. 1000BASE-T full duplex 1000BASE-T 100BASE-T2 full duplex 100BASE-TX full duplex 100BASE-T2 100BASE-T4 100BASE-TX 10BASE-T full duplex 10BAST-T However PHY drivers didn't honor the order such that 100BASE-T4 had higher priority than 100BASE-TX full duplex. Fix that long standing bugs such that have PHY drivers choose the highest common denominator ability. Fix a bug in dcphy which inadvertently aceepts 100BASE-T4. PR: 92599
2007-11-16 10:25:36 +00:00
if (anlpar & ANLPAR_TX_FD)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
ifmr->ifm_active |= IFM_100_TX|IFM_FDX;
IEEE 802.3 Annex 28B.3 explicitly specifies the following relative priorities of the technologies supported by 802.3 Selector Field value. 1000BASE-T full duplex 1000BASE-T 100BASE-T2 full duplex 100BASE-TX full duplex 100BASE-T2 100BASE-T4 100BASE-TX 10BASE-T full duplex 10BAST-T However PHY drivers didn't honor the order such that 100BASE-T4 had higher priority than 100BASE-TX full duplex. Fix that long standing bugs such that have PHY drivers choose the highest common denominator ability. Fix a bug in dcphy which inadvertently aceepts 100BASE-T4. PR: 92599
2007-11-16 10:25:36 +00:00
else if (anlpar & ANLPAR_T4)
ifmr->ifm_active |= IFM_100_T4;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
else if (anlpar & ANLPAR_TX)
ifmr->ifm_active |= IFM_100_TX;
else if (anlpar & ANLPAR_10_FD)
ifmr->ifm_active |= IFM_10_T|IFM_FDX;
else if (anlpar & ANLPAR_10)
ifmr->ifm_active |= IFM_10_T;
else
ifmr->ifm_active |= IFM_NONE;
} else
ifmr->ifm_active = ifm->ifm_cur->ifm_media;
return;
bail:
- Add locking and mark MPSAFE. The driver had a mutex in the softc and even initialized it, but it never used it. - Use callout_*() to manage the callout. - Use m_devget() to copy data out of the rx buffers rather than doing it all by hand. - Use m_getcl() to allocate mbuf clusters rather than doing it all by hand. - Don't free the software descriptor for a rx ring entry if we can't allocate an mbuf cluster for it. We left a dangling pointer and never reallocated the entry anyway. OpenBSD's code (from which this was derived) has the same bug. Tested by: NO ONE (despite repeated requests) Reviewed by: wpaul (5)
2005-10-27 21:16:17 +00:00
TXP_UNLOCK(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
ifmr->ifm_active |= IFM_NONE;
ifmr->ifm_status &= ~IFM_AVALID;
}
#ifdef TXP_DEBUG
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_show_descriptor(void *d)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
struct txp_cmd_desc *cmd = d;
struct txp_rsp_desc *rsp = d;
struct txp_tx_desc *txd = d;
struct txp_frag_desc *frgd = d;
switch (cmd->cmd_flags & CMD_FLAGS_TYPE_M) {
case CMD_FLAGS_TYPE_CMD:
/* command descriptor */
printf("[cmd flags 0x%x num %d id %d seq %d par1 0x%x par2 0x%x par3 0x%x]\n",
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
cmd->cmd_flags, cmd->cmd_numdesc, le16toh(cmd->cmd_id),
le16toh(cmd->cmd_seq), le16toh(cmd->cmd_par1),
le32toh(cmd->cmd_par2), le32toh(cmd->cmd_par3));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
case CMD_FLAGS_TYPE_RESP:
/* response descriptor */
printf("[rsp flags 0x%x num %d id %d seq %d par1 0x%x par2 0x%x par3 0x%x]\n",
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
rsp->rsp_flags, rsp->rsp_numdesc, le16toh(rsp->rsp_id),
le16toh(rsp->rsp_seq), le16toh(rsp->rsp_par1),
le32toh(rsp->rsp_par2), le32toh(rsp->rsp_par3));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
case CMD_FLAGS_TYPE_DATA:
/* data header (assuming tx for now) */
printf("[data flags 0x%x num %d totlen %d addr 0x%x/0x%x pflags 0x%x]",
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txd->tx_flags, txd->tx_numdesc, le16toh(txd->tx_totlen),
le32toh(txd->tx_addrlo), le32toh(txd->tx_addrhi),
le32toh(txd->tx_pflags));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
case CMD_FLAGS_TYPE_FRAG:
/* fragment descriptor */
printf("[frag flags 0x%x rsvd1 0x%x len %d addr 0x%x/0x%x rsvd2 0x%x]",
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
frgd->frag_flags, frgd->frag_rsvd1, le16toh(frgd->frag_len),
le32toh(frgd->frag_addrlo), le32toh(frgd->frag_addrhi),
le32toh(frgd->frag_rsvd2));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
default:
printf("[unknown(%x) flags 0x%x num %d id %d seq %d par1 0x%x par2 0x%x par3 0x%x]\n",
cmd->cmd_flags & CMD_FLAGS_TYPE_M,
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
cmd->cmd_flags, cmd->cmd_numdesc, le16toh(cmd->cmd_id),
le16toh(cmd->cmd_seq), le16toh(cmd->cmd_par1),
le32toh(cmd->cmd_par2), le32toh(cmd->cmd_par3));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
break;
}
}
#endif
static void
K&R -> ANSI C function definitions.
2009-03-11 08:25:18 +00:00
txp_set_filter(struct txp_softc *sc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
Replace local CRC32 routine with ether_crc32_be(). This should have happened long time ago. Also simplify Rx filter logic.
2009-03-11 09:57:11 +00:00
struct ifnet *ifp;
uint32_t crc, mchash[2];
s/u_int8_t/uint8_t/g s/u_int16_t/uint16_t/g s/u_int32_t/uint32_t/g s/u_int64_t/uint64_t/g
2009-03-11 09:06:39 +00:00
uint16_t filter;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
struct ifmultiaddr *ifma;
Replace local CRC32 routine with ether_crc32_be(). This should have happened long time ago. Also simplify Rx filter logic.
2009-03-11 09:57:11 +00:00
int mcnt;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
Replace local CRC32 routine with ether_crc32_be(). This should have happened long time ago. Also simplify Rx filter logic.
2009-03-11 09:57:11 +00:00
TXP_LOCK_ASSERT(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
Replace local CRC32 routine with ether_crc32_be(). This should have happened long time ago. Also simplify Rx filter logic.
2009-03-11 09:57:11 +00:00
ifp = sc->sc_ifp;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
filter = TXP_RXFILT_DIRECT;
Replace local CRC32 routine with ether_crc32_be(). This should have happened long time ago. Also simplify Rx filter logic.
2009-03-11 09:57:11 +00:00
if ((ifp->if_flags & IFF_BROADCAST) != 0)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
filter |= TXP_RXFILT_BROADCAST;
Replace local CRC32 routine with ether_crc32_be(). This should have happened long time ago. Also simplify Rx filter logic.
2009-03-11 09:57:11 +00:00
if ((ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
if ((ifp->if_flags & IFF_ALLMULTI) != 0)
filter |= TXP_RXFILT_ALLMULTI;
if ((ifp->if_flags & IFF_PROMISC) != 0)
filter = TXP_RXFILT_PROMISC;
goto setit;
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
Replace local CRC32 routine with ether_crc32_be(). This should have happened long time ago. Also simplify Rx filter logic.
2009-03-11 09:57:11 +00:00
mchash[0] = mchash[1] = 0;
mcnt = 0;
Use if_maddr_rlock()/if_maddr_runlock() rather than IF_ADDR_LOCK()/ IF_ADDR_UNLOCK() across network device drivers when accessing the per-interface multicast address list, if_multiaddrs. This will allow us to change the locking strategy without affecting our driver programming interface or binary interface. For two wireless drivers, remove unnecessary locking, since they don't actually access the multicast address list. Approved by: re (kib) MFC after: 6 weeks
2009-06-26 11:45:06 +00:00
if_maddr_rlock(ifp);
Replace local CRC32 routine with ether_crc32_be(). This should have happened long time ago. Also simplify Rx filter logic.
2009-03-11 09:57:11 +00:00
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
crc = ether_crc32_be(LLADDR((struct sockaddr_dl *)
ifma->ifma_addr), ETHER_ADDR_LEN);
crc &= 0x3f;
mchash[crc >> 5] |= 1 << (crc & 0x1f);
mcnt++;
}
Use if_maddr_rlock()/if_maddr_runlock() rather than IF_ADDR_LOCK()/ IF_ADDR_UNLOCK() across network device drivers when accessing the per-interface multicast address list, if_multiaddrs. This will allow us to change the locking strategy without affecting our driver programming interface or binary interface. For two wireless drivers, remove unnecessary locking, since they don't actually access the multicast address list. Approved by: re (kib) MFC after: 6 weeks
2009-06-26 11:45:06 +00:00
if_maddr_runlock(ifp);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
Replace local CRC32 routine with ether_crc32_be(). This should have happened long time ago. Also simplify Rx filter logic.
2009-03-11 09:57:11 +00:00
if (mcnt > 0) {
filter |= TXP_RXFILT_HASHMULTI;
txp_command(sc, TXP_CMD_MCAST_HASH_MASK_WRITE, 2, mchash[0],
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
mchash[1], NULL, NULL, NULL, TXP_CMD_NOWAIT);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
setit:
txp_command(sc, TXP_CMD_RX_FILTER_WRITE, filter, 0, 0,
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
NULL, NULL, NULL, TXP_CMD_NOWAIT);
}
static int
txp_set_capabilities(struct txp_softc *sc)
{
struct ifnet *ifp;
uint32_t rxcap, txcap;
TXP_LOCK_ASSERT(sc);
rxcap = txcap = 0;
ifp = sc->sc_ifp;
if ((ifp->if_capenable & IFCAP_TXCSUM) != 0) {
if ((ifp->if_hwassist & CSUM_IP) != 0)
txcap |= OFFLOAD_IPCKSUM;
if ((ifp->if_hwassist & CSUM_TCP) != 0)
txcap |= OFFLOAD_TCPCKSUM;
if ((ifp->if_hwassist & CSUM_UDP) != 0)
txcap |= OFFLOAD_UDPCKSUM;
rxcap = txcap;
}
if ((ifp->if_capenable & IFCAP_RXCSUM) == 0)
rxcap &= ~(OFFLOAD_IPCKSUM | OFFLOAD_TCPCKSUM |
OFFLOAD_UDPCKSUM);
if ((ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
rxcap |= OFFLOAD_VLAN;
txcap |= OFFLOAD_VLAN;
}
/* Tell firmware new offload configuration. */
return (txp_command(sc, TXP_CMD_OFFLOAD_WRITE, 0, txcap, rxcap, NULL,
NULL, NULL, TXP_CMD_NOWAIT));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
static void
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
txp_stats_save(struct txp_softc *sc)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
{
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
struct txp_rsp_desc *rsp;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_LOCK_ASSERT(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
rsp = NULL;
if (txp_ext_command(sc, TXP_CMD_READ_STATISTICS, 0, 0, 0, NULL, 0,
&rsp, TXP_CMD_WAIT))
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
goto out;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (rsp->rsp_numdesc != 6)
goto out;
txp_stats_update(sc, rsp);
out:
if (rsp != NULL)
free(rsp, M_DEVBUF);
bcopy(&sc->sc_stats, &sc->sc_ostats, sizeof(struct txp_hw_stats));
}
static void
txp_stats_update(struct txp_softc *sc, struct txp_rsp_desc *rsp)
{
struct txp_hw_stats *ostats, *stats;
struct txp_ext_desc *ext;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
TXP_LOCK_ASSERT(sc);
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
ext = (struct txp_ext_desc *)(rsp + 1);
ostats = &sc->sc_ostats;
stats = &sc->sc_stats;
stats->tx_frames = ostats->tx_frames + le32toh(rsp->rsp_par2);
stats->tx_bytes = ostats->tx_bytes + (uint64_t)le32toh(rsp->rsp_par3) +
((uint64_t)le32toh(ext[0].ext_1) << 32);
stats->tx_deferred = ostats->tx_deferred + le32toh(ext[0].ext_2);
stats->tx_late_colls = ostats->tx_late_colls + le32toh(ext[0].ext_3);
stats->tx_colls = ostats->tx_colls + le32toh(ext[0].ext_4);
stats->tx_carrier_lost = ostats->tx_carrier_lost +
le32toh(ext[1].ext_1);
stats->tx_multi_colls = ostats->tx_multi_colls +
le32toh(ext[1].ext_2);
stats->tx_excess_colls = ostats->tx_excess_colls +
le32toh(ext[1].ext_3);
stats->tx_fifo_underruns = ostats->tx_fifo_underruns +
le32toh(ext[1].ext_4);
stats->tx_mcast_oflows = ostats->tx_mcast_oflows +
le32toh(ext[2].ext_1);
stats->tx_filtered = ostats->tx_filtered + le32toh(ext[2].ext_2);
stats->rx_frames = ostats->rx_frames + le32toh(ext[2].ext_3);
stats->rx_bytes = ostats->rx_bytes + (uint64_t)le32toh(ext[2].ext_4) +
((uint64_t)le32toh(ext[3].ext_1) << 32);
stats->rx_fifo_oflows = ostats->rx_fifo_oflows + le32toh(ext[3].ext_2);
stats->rx_badssd = ostats->rx_badssd + le32toh(ext[3].ext_3);
stats->rx_crcerrs = ostats->rx_crcerrs + le32toh(ext[3].ext_4);
stats->rx_lenerrs = ostats->rx_lenerrs + le32toh(ext[4].ext_1);
stats->rx_bcast_frames = ostats->rx_bcast_frames +
le32toh(ext[4].ext_2);
stats->rx_mcast_frames = ostats->rx_mcast_frames +
le32toh(ext[4].ext_3);
stats->rx_oflows = ostats->rx_oflows + le32toh(ext[4].ext_4);
stats->rx_filtered = ostats->rx_filtered + le32toh(ext[5].ext_1);
- Provide txp_get_counter() to return counters that are not collected, but taken from hardware. - Mechanically convert to if_inc_counter() the rest of counters.
2014-09-24 11:58:23 +00:00
}
static uint64_t
txp_get_counter(struct ifnet *ifp, ift_counter cnt)
{
struct txp_softc *sc;
struct txp_hw_stats *stats;
sc = if_getsoftc(ifp);
stats = &sc->sc_stats;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
- Provide txp_get_counter() to return counters that are not collected, but taken from hardware. - Mechanically convert to if_inc_counter() the rest of counters.
2014-09-24 11:58:23 +00:00
switch (cnt) {
case IFCOUNTER_IERRORS:
return (stats->rx_fifo_oflows + stats->rx_badssd +
stats->rx_crcerrs + stats->rx_lenerrs + stats->rx_oflows);
case IFCOUNTER_OERRORS:
return (stats->tx_deferred + stats->tx_carrier_lost +
stats->tx_fifo_underruns + stats->tx_mcast_oflows);
case IFCOUNTER_COLLISIONS:
return (stats->tx_late_colls + stats->tx_multi_colls +
stats->tx_excess_colls);
case IFCOUNTER_OPACKETS:
return (stats->tx_frames);
case IFCOUNTER_IPACKETS:
return (stats->rx_frames);
default:
return (if_get_counter_default(ifp, cnt));
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#define TXP_SYSCTL_STAT_ADD32(c, h, n, p, d) \
SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the rest of the devices.
2011-01-12 19:53:56 +00:00
#if __FreeBSD_version >= 900030
#define TXP_SYSCTL_STAT_ADD64(c, h, n, p, d) \
SYSCTL_ADD_UQUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
#elif __FreeBSD_version > 800000
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#define TXP_SYSCTL_STAT_ADD64(c, h, n, p, d) \
SYSCTL_ADD_QUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
#else
#define TXP_SYSCTL_STAT_ADD64(c, h, n, p, d) \
SYSCTL_ADD_ULONG(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
#endif
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static void
txp_sysctl_node(struct txp_softc *sc)
{
struct sysctl_ctx_list *ctx;
struct sysctl_oid_list *child, *parent;
struct sysctl_oid *tree;
struct txp_hw_stats *stats;
int error;
stats = &sc->sc_stats;
ctx = device_get_sysctl_ctx(sc->sc_dev);
child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->sc_dev));
SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "process_limit",
CTLTYPE_INT | CTLFLAG_RW, &sc->sc_process_limit, 0,
sysctl_hw_txp_proc_limit, "I",
"max number of Rx events to process");
/* Pull in device tunables. */
sc->sc_process_limit = TXP_PROC_DEFAULT;
error = resource_int_value(device_get_name(sc->sc_dev),
device_get_unit(sc->sc_dev), "process_limit",
&sc->sc_process_limit);
if (error == 0) {
if (sc->sc_process_limit < TXP_PROC_MIN ||
sc->sc_process_limit > TXP_PROC_MAX) {
device_printf(sc->sc_dev,
"process_limit value out of range; "
"using default: %d\n", TXP_PROC_DEFAULT);
sc->sc_process_limit = TXP_PROC_DEFAULT;
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats", CTLFLAG_RD,
NULL, "TXP statistics");
parent = SYSCTL_CHILDREN(tree);
/* Tx statistics. */
tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD,
NULL, "Tx MAC statistics");
child = SYSCTL_CHILDREN(tree);
TXP_SYSCTL_STAT_ADD32(ctx, child, "frames",
&stats->tx_frames, "Frames");
TXP_SYSCTL_STAT_ADD64(ctx, child, "octets",
&stats->tx_bytes, "Octets");
TXP_SYSCTL_STAT_ADD32(ctx, child, "deferred",
&stats->tx_deferred, "Deferred frames");
TXP_SYSCTL_STAT_ADD32(ctx, child, "late_colls",
&stats->tx_late_colls, "Late collisions");
TXP_SYSCTL_STAT_ADD32(ctx, child, "colls",
&stats->tx_colls, "Collisions");
TXP_SYSCTL_STAT_ADD32(ctx, child, "carrier_lost",
&stats->tx_carrier_lost, "Carrier lost");
TXP_SYSCTL_STAT_ADD32(ctx, child, "multi_colls",
&stats->tx_multi_colls, "Multiple collisions");
TXP_SYSCTL_STAT_ADD32(ctx, child, "excess_colls",
&stats->tx_excess_colls, "Excessive collisions");
TXP_SYSCTL_STAT_ADD32(ctx, child, "fifo_underruns",
&stats->tx_fifo_underruns, "FIFO underruns");
TXP_SYSCTL_STAT_ADD32(ctx, child, "mcast_oflows",
&stats->tx_mcast_oflows, "Multicast overflows");
TXP_SYSCTL_STAT_ADD32(ctx, child, "filtered",
&stats->tx_filtered, "Filtered frames");
/* Rx statistics. */
tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD,
NULL, "Rx MAC statistics");
child = SYSCTL_CHILDREN(tree);
TXP_SYSCTL_STAT_ADD32(ctx, child, "frames",
&stats->rx_frames, "Frames");
TXP_SYSCTL_STAT_ADD64(ctx, child, "octets",
&stats->rx_bytes, "Octets");
TXP_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows",
&stats->rx_fifo_oflows, "FIFO overflows");
TXP_SYSCTL_STAT_ADD32(ctx, child, "badssd",
&stats->rx_badssd, "Bad SSD");
TXP_SYSCTL_STAT_ADD32(ctx, child, "crcerrs",
&stats->rx_crcerrs, "CRC errors");
TXP_SYSCTL_STAT_ADD32(ctx, child, "lenerrs",
&stats->rx_lenerrs, "Length errors");
TXP_SYSCTL_STAT_ADD32(ctx, child, "bcast_frames",
&stats->rx_bcast_frames, "Broadcast frames");
TXP_SYSCTL_STAT_ADD32(ctx, child, "mcast_frames",
&stats->rx_mcast_frames, "Multicast frames");
TXP_SYSCTL_STAT_ADD32(ctx, child, "oflows",
&stats->rx_oflows, "Overflows");
TXP_SYSCTL_STAT_ADD32(ctx, child, "filtered",
&stats->rx_filtered, "Filtered frames");
}
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
#undef TXP_SYSCTL_STAT_ADD32
#undef TXP_SYSCTL_STAT_ADD64
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
static int
sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
{
int error, value;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
if (arg1 == NULL)
return (EINVAL);
value = *(int *)arg1;
error = sysctl_handle_int(oidp, &value, 0, req);
if (error || req->newptr == NULL)
return (error);
if (value < low || value > high)
return (EINVAL);
*(int *)arg1 = value;
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
return (0);
}
static int
sysctl_hw_txp_proc_limit(SYSCTL_HANDLER_ARGS)
{
return (sysctl_int_range(oidp, arg1, arg2, req,
TXP_PROC_MIN, TXP_PROC_MAX));
You were knocked senseless by the Boomerang, spun around by the Cyclone, blown over by the Hurricane and had a house dropped on you by the Tornado. Now it's time to have your parade rained on by... the Typhoon! This commit adds driver support for 3Com 3cR990 10/100 ethernet adapters based on the Typhoon I and Typhoon II chipsets. This is actually a port of the OpenBSD driver with many hacks by me. No Virginia, there isn't any support for the hardware crypto yet. However there is support for TCP/IP checksum offload and VLANs. Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for squeezing enough info out of 3Com to get this written, and for doing most of the hard work. Manual page is included. Compiled as a module and included in GENERIC.
2001-07-23 20:44:54 +00:00
}
Reference in New Issue Copy Permalink