freebsd-skq/sys/dev/txp/if_txpreg.h

769 lines
25 KiB
C
Raw Normal View History

/* $OpenBSD: if_txpreg.h,v 1.30 2001/06/23 04:18:02 jason Exp $ */
/* $FreeBSD$ */
/*-
* Copyright (c) 2001 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 Aaron Campbell.
* 4. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 THE AUTHOR OR HIS RELATIVES 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 MIND, 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.
*/
/*
* Typhoon registers.
*/
#define TXP_SRR 0x00 /* soft reset register */
#define TXP_ISR 0x04 /* interrupt status register */
#define TXP_IER 0x08 /* interrupt enable register */
#define TXP_IMR 0x0c /* interrupt mask register */
#define TXP_SIR 0x10 /* self interrupt register */
#define TXP_H2A_7 0x14 /* host->arm comm 7 */
#define TXP_H2A_6 0x18 /* host->arm comm 6 */
#define TXP_H2A_5 0x1c /* host->arm comm 5 */
#define TXP_H2A_4 0x20 /* host->arm comm 4 */
#define TXP_H2A_3 0x24 /* host->arm comm 3 */
#define TXP_H2A_2 0x28 /* host->arm comm 2 */
#define TXP_H2A_1 0x2c /* host->arm comm 1 */
#define TXP_H2A_0 0x30 /* host->arm comm 0 */
#define TXP_A2H_3 0x34 /* arm->host comm 3 */
#define TXP_A2H_2 0x38 /* arm->host comm 2 */
#define TXP_A2H_1 0x3c /* arm->host comm 1 */
#define TXP_A2H_0 0x40 /* arm->host comm 0 */
/*
* interrupt bits (IMR, ISR, IER)
*/
#define TXP_INT_RESERVED 0xffff0000
#define TXP_INT_A2H_7 0x00008000 /* arm->host comm 7 */
#define TXP_INT_A2H_6 0x00004000 /* arm->host comm 6 */
#define TXP_INT_A2H_5 0x00002000 /* arm->host comm 5 */
#define TXP_INT_A2H_4 0x00001000 /* arm->host comm 4 */
#define TXP_INT_SELF 0x00000800 /* self interrupt */
#define TXP_INT_PCI_TABORT 0x00000400 /* pci target abort */
#define TXP_INT_PCI_MABORT 0x00000200 /* pci master abort */
#define TXP_INT_DMA3 0x00000100 /* dma3 done */
#define TXP_INT_DMA2 0x00000080 /* dma2 done */
#define TXP_INT_DMA1 0x00000040 /* dma1 done */
#define TXP_INT_DMA0 0x00000020 /* dma0 done */
#define TXP_INT_A2H_3 0x00000010 /* arm->host comm 3 */
#define TXP_INT_A2H_2 0x00000008 /* arm->host comm 2 */
#define TXP_INT_A2H_1 0x00000004 /* arm->host comm 1 */
#define TXP_INT_A2H_0 0x00000002 /* arm->host comm 0 */
#define TXP_INT_LATCH 0x00000001 /* interrupt latch */
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
/*
* Controller periodically generates TXP_INT_A2H_3 interrupt so
* we don't want to see them in interrupt handler.
*/
#define TXP_INTRS 0xFFFFFFEF
#define TXP_INTR_ALL 0xFFFFFFFF
#define TXP_INTR_NONE 0x00000000
/*
* soft reset register (SRR)
*/
#define TXP_SRR_ALL 0x0000007f /* full reset */
/*
* Typhoon boot commands.
*/
#define TXP_BOOTCMD_NULL 0x00
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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_BOOTCMD_WAKEUP 0xfa
#define TXP_BOOTCMD_DOWNLOAD_COMPLETE 0xfb
#define TXP_BOOTCMD_SEGMENT_AVAILABLE 0xfc
#define TXP_BOOTCMD_RUNTIME_IMAGE 0xfd
#define TXP_BOOTCMD_REGISTER_BOOT_RECORD 0xff
/*
* Typhoon runtime commands.
*/
#define TXP_CMD_GLOBAL_RESET 0x00
#define TXP_CMD_TX_ENABLE 0x01
#define TXP_CMD_TX_DISABLE 0x02
#define TXP_CMD_RX_ENABLE 0x03
#define TXP_CMD_RX_DISABLE 0x04
#define TXP_CMD_RX_FILTER_WRITE 0x05
#define TXP_CMD_RX_FILTER_READ 0x06
#define TXP_CMD_READ_STATISTICS 0x07
#define TXP_CMD_CYCLE_STATISTICS 0x08
#define TXP_CMD_CLEAR_STATISTICS 0x09
#define TXP_CMD_MEMORY_READ 0x0a
#define TXP_CMD_MEMORY_WRITE_SINGLE 0x0b
#define TXP_CMD_VARIABLE_SECTION_READ 0x0c
#define TXP_CMD_VARIABLE_SECTION_WRITE 0x0d
#define TXP_CMD_STATIC_SECTION_READ 0x0e
#define TXP_CMD_STATIC_SECTION_WRITE 0x0f
#define TXP_CMD_IMAGE_SECTION_PROGRAM 0x10
#define TXP_CMD_NVRAM_PAGE_READ 0x11
#define TXP_CMD_NVRAM_PAGE_WRITE 0x12
#define TXP_CMD_XCVR_SELECT 0x13
#define TXP_CMD_TEST_MUX 0x14
#define TXP_CMD_PHYLOOPBACK_ENABLE 0x15
#define TXP_CMD_PHYLOOPBACK_DISABLE 0x16
#define TXP_CMD_MAC_CONTROL_READ 0x17
#define TXP_CMD_MAC_CONTROL_WRITE 0x18
#define TXP_CMD_MAX_PKT_SIZE_READ 0x19
#define TXP_CMD_MAX_PKT_SIZE_WRITE 0x1a
#define TXP_CMD_MEDIA_STATUS_READ 0x1b
#define TXP_CMD_MEDIA_STATUS_WRITE 0x1c
#define TXP_CMD_NETWORK_DIAGS_READ 0x1d
#define TXP_CMD_NETWORK_DIAGS_WRITE 0x1e
#define TXP_CMD_PHY_MGMT_READ 0x1f
#define TXP_CMD_PHY_MGMT_WRITE 0x20
#define TXP_CMD_VARIABLE_PARAMETER_READ 0x21
#define TXP_CMD_VARIABLE_PARAMETER_WRITE 0x22
#define TXP_CMD_GOTO_SLEEP 0x23
#define TXP_CMD_FIREWALL_CONTROL 0x24
#define TXP_CMD_MCAST_HASH_MASK_WRITE 0x25
#define TXP_CMD_STATION_ADDRESS_WRITE 0x26
#define TXP_CMD_STATION_ADDRESS_READ 0x27
#define TXP_CMD_STATION_MASK_WRITE 0x28
#define TXP_CMD_STATION_MASK_READ 0x29
#define TXP_CMD_VLAN_ETHER_TYPE_READ 0x2a
#define TXP_CMD_VLAN_ETHER_TYPE_WRITE 0x2b
#define TXP_CMD_VLAN_MASK_READ 0x2c
#define TXP_CMD_VLAN_MASK_WRITE 0x2d
#define TXP_CMD_BCAST_THROTTLE_WRITE 0x2e
#define TXP_CMD_BCAST_THROTTLE_READ 0x2f
#define TXP_CMD_DHCP_PREVENT_WRITE 0x30
#define TXP_CMD_DHCP_PREVENT_READ 0x31
#define TXP_CMD_RECV_BUFFER_CONTROL 0x32
#define TXP_CMD_SOFTWARE_RESET 0x33
#define TXP_CMD_CREATE_SA 0x34
#define TXP_CMD_DELETE_SA 0x35
#define TXP_CMD_ENABLE_RX_IP_OPTION 0x36
#define TXP_CMD_RANDOM_NUMBER_CONTROL 0x37
#define TXP_CMD_RANDOM_NUMBER_READ 0x38
#define TXP_CMD_MATRIX_TABLE_MODE_WRITE 0x39
#define TXP_CMD_MATRIX_DETAIL_READ 0x3a
#define TXP_CMD_FILTER_ARRAY_READ 0x3b
#define TXP_CMD_FILTER_DETAIL_READ 0x3c
#define TXP_CMD_FILTER_TABLE_MODE_WRITE 0x3d
#define TXP_CMD_FILTER_TCL_WRITE 0x3e
#define TXP_CMD_FILTER_TBL_READ 0x3f
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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_CMD_READ_VERSION 0x43
#define TXP_CMD_FILTER_DEFINE 0x45
#define TXP_CMD_ADD_WAKEUP_PKT 0x46
#define TXP_CMD_ADD_SLEEP_PKT 0x47
#define TXP_CMD_ENABLE_SLEEP_EVENTS 0x48
#define TXP_CMD_ENABLE_WAKEUP_EVENTS 0x49
#define TXP_CMD_GET_IP_ADDRESS 0x4a
#define TXP_CMD_READ_PCI_REG 0x4c
#define TXP_CMD_WRITE_PCI_REG 0x4d
#define TXP_CMD_OFFLOAD_READ 0x4e
#define TXP_CMD_OFFLOAD_WRITE 0x4f
#define TXP_CMD_HELLO_RESPONSE 0x57
#define TXP_CMD_ENABLE_RX_FILTER 0x58
#define TXP_CMD_RX_FILTER_CAPABILITY 0x59
#define TXP_CMD_HALT 0x5d
#define TXP_CMD_READ_IPSEC_INFO 0x54
#define TXP_CMD_GET_IPSEC_ENABLE 0x67
#define TXP_CMD_INVALID 0xffff
#define TXP_FRAGMENT 0x0000
#define TXP_TXFRAME 0x0001
#define TXP_COMMAND 0x0002
#define TXP_OPTION 0x0003
#define TXP_RECEIVE 0x0004
#define TXP_RESPONSE 0x0005
#define TXP_TYPE_IPSEC 0x0000
#define TXP_TYPE_TCPSEGMENT 0x0001
#define TXP_PFLAG_NOCRC 0x0000
#define TXP_PFLAG_IPCKSUM 0x0001
#define TXP_PFLAG_TCPCKSUM 0x0002
#define TXP_PFLAG_TCPSEGMENT 0x0004
#define TXP_PFLAG_INSERTVLAN 0x0008
#define TXP_PFLAG_IPSEC 0x0010
#define TXP_PFLAG_PRIORITY 0x0020
#define TXP_PFLAG_UDPCKSUM 0x0040
#define TXP_PFLAG_PADFRAME 0x0080
#define TXP_MISC_FIRSTDESC 0x0000
#define TXP_MISC_LASTDESC 0x0001
#define TXP_ERR_INTERNAL 0x0000
#define TXP_ERR_FIFOUNDERRUN 0x0001
#define TXP_ERR_BADSSD 0x0002
#define TXP_ERR_RUNT 0x0003
#define TXP_ERR_CRC 0x0004
#define TXP_ERR_OVERSIZE 0x0005
#define TXP_ERR_ALIGNMENT 0x0006
#define TXP_ERR_DRIBBLEBIT 0x0007
#define TXP_PROTO_UNKNOWN 0x0000
#define TXP_PROTO_IP 0x0001
#define TXP_PROTO_IPX 0x0002
#define TXP_PROTO_RESERVED 0x0003
#define TXP_STAT_PROTO 0x0001
#define TXP_STAT_VLAN 0x0002
#define TXP_STAT_IPFRAGMENT 0x0004
#define TXP_STAT_IPSEC 0x0008
#define TXP_STAT_IPCKSUMBAD 0x0010
#define TXP_STAT_TCPCKSUMBAD 0x0020
#define TXP_STAT_UDPCKSUMBAD 0x0040
#define TXP_STAT_IPCKSUMGOOD 0x0080
#define TXP_STAT_TCPCKSUMGOOD 0x0100
#define TXP_STAT_UDPCKSUMGOOD 0x0200
struct txp_tx_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
uint8_t tx_flags; /* type/descriptor flags */
uint8_t tx_numdesc; /* number of descriptors */
uint16_t tx_totlen; /* total packet length */
uint32_t tx_addrlo; /* virt addr low word */
uint32_t tx_addrhi; /* virt addr high word */
uint32_t tx_pflags; /* processing flags */
};
#define TX_FLAGS_TYPE_M 0x07 /* type mask */
#define TX_FLAGS_TYPE_FRAG 0x00 /* type: fragment */
#define TX_FLAGS_TYPE_DATA 0x01 /* type: data frame */
#define TX_FLAGS_TYPE_CMD 0x02 /* type: command frame */
#define TX_FLAGS_TYPE_OPT 0x03 /* type: options */
#define TX_FLAGS_TYPE_RX 0x04 /* type: command */
#define TX_FLAGS_TYPE_RESP 0x05 /* type: response */
#define TX_FLAGS_RESP 0x40 /* response requested */
#define TX_FLAGS_VALID 0x80 /* valid descriptor */
#define TX_PFLAGS_DNAC 0x00000001 /* do not add crc */
#define TX_PFLAGS_IPCKSUM 0x00000002 /* ip checksum */
#define TX_PFLAGS_TCPCKSUM 0x00000004 /* tcp checksum */
#define TX_PFLAGS_TCPSEG 0x00000008 /* tcp segmentation */
#define TX_PFLAGS_VLAN 0x00000010 /* insert vlan */
#define TX_PFLAGS_IPSEC 0x00000020 /* perform ipsec */
#define TX_PFLAGS_PRIO 0x00000040 /* priority field valid */
#define TX_PFLAGS_UDPCKSUM 0x00000080 /* udp checksum */
#define TX_PFLAGS_PADFRAME 0x00000100 /* pad frame */
#define TX_PFLAGS_VLANTAG_M 0x0ffff000 /* vlan tag mask */
#define TX_PFLAGS_VLANPRI_M 0x00700000 /* vlan priority mask */
#define TX_PFLAGS_VLANTAG_S 12 /* amount to shift tag */
struct txp_rx_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
uint8_t rx_flags; /* type/descriptor flags */
uint8_t rx_numdesc; /* number of descriptors */
uint16_t rx_len; /* frame length */
uint32_t rx_vaddrlo; /* virtual address, lo word */
uint32_t rx_vaddrhi; /* virtual address, hi word */
uint32_t rx_stat; /* status */
uint16_t rx_filter; /* filter status */
uint16_t rx_hash; /* hash status */
uint32_t rx_vlan; /* vlan tag/priority */
};
/* txp_rx_desc.rx_flags */
#define RX_FLAGS_TYPE_M 0x07 /* type mask */
#define RX_FLAGS_TYPE_FRAG 0x00 /* type: fragment */
#define RX_FLAGS_TYPE_DATA 0x01 /* type: data frame */
#define RX_FLAGS_TYPE_CMD 0x02 /* type: command frame */
#define RX_FLAGS_TYPE_OPT 0x03 /* type: options */
#define RX_FLAGS_TYPE_RX 0x04 /* type: command */
#define RX_FLAGS_TYPE_RESP 0x05 /* type: response */
#define RX_FLAGS_RCV_TYPE_M 0x18 /* rcvtype mask */
#define RX_FLAGS_RCV_TYPE_RX 0x00 /* rcvtype: receive */
#define RX_FLAGS_RCV_TYPE_RSP 0x08 /* rcvtype: response */
#define RX_FLAGS_ERROR 0x40 /* error in packet */
/* txp_rx_desc.rx_stat (if rx_flags & RX_FLAGS_ERROR bit set) */
#define RX_ERROR_ADAPTER 0x00000000 /* adapter internal error */
#define RX_ERROR_FIFO 0x00000001 /* fifo underrun */
#define RX_ERROR_BADSSD 0x00000002 /* bad ssd */
#define RX_ERROR_RUNT 0x00000003 /* runt packet */
#define RX_ERROR_CRC 0x00000004 /* bad crc */
#define RX_ERROR_OVERSIZE 0x00000005 /* oversized packet */
#define RX_ERROR_ALIGN 0x00000006 /* alignment error */
#define RX_ERROR_DRIBBLE 0x00000007 /* dribble bit */
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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 RX_ERROR_MASK 0x07
/* txp_rx_desc.rx_stat (if rx_flags & RX_FLAGS_ERROR not bit set) */
#define RX_STAT_PROTO_M 0x00000003 /* protocol mask */
#define RX_STAT_PROTO_UK 0x00000000 /* unknown protocol */
#define RX_STAT_PROTO_IPX 0x00000001 /* IPX */
#define RX_STAT_PROTO_IP 0x00000002 /* IP */
#define RX_STAT_PROTO_RSV 0x00000003 /* reserved */
#define RX_STAT_VLAN 0x00000004 /* vlan tag (in rxd) */
#define RX_STAT_IPFRAG 0x00000008 /* fragment, ipsec not done */
#define RX_STAT_IPSEC 0x00000010 /* ipsec decoded packet */
#define RX_STAT_IPCKSUMBAD 0x00000020 /* ip checksum failed */
#define RX_STAT_UDPCKSUMBAD 0x00000040 /* udp checksum failed */
#define RX_STAT_TCPCKSUMBAD 0x00000080 /* tcp checksum failed */
#define RX_STAT_IPCKSUMGOOD 0x00000100 /* ip checksum succeeded */
#define RX_STAT_UDPCKSUMGOOD 0x00000200 /* udp checksum succeeded */
#define RX_STAT_TCPCKSUMGOOD 0x00000400 /* tcp checksum succeeded */
struct txp_rxbuf_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
uint32_t rb_paddrlo;
uint32_t rb_paddrhi;
uint32_t rb_vaddrlo;
uint32_t rb_vaddrhi;
};
/* Extension descriptor */
struct txp_ext_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
uint32_t ext_1;
uint32_t ext_2;
uint32_t ext_3;
uint32_t ext_4;
};
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
uint8_t cmd_flags;
uint8_t cmd_numdesc;
uint16_t cmd_id;
uint16_t cmd_seq;
uint16_t cmd_par1;
uint32_t cmd_par2;
uint32_t cmd_par3;
};
#define CMD_FLAGS_TYPE_M 0x07 /* type mask */
#define CMD_FLAGS_TYPE_FRAG 0x00 /* type: fragment */
#define CMD_FLAGS_TYPE_DATA 0x01 /* type: data frame */
#define CMD_FLAGS_TYPE_CMD 0x02 /* type: command frame */
#define CMD_FLAGS_TYPE_OPT 0x03 /* type: options */
#define CMD_FLAGS_TYPE_RX 0x04 /* type: command */
#define CMD_FLAGS_TYPE_RESP 0x05 /* type: response */
#define CMD_FLAGS_RESP 0x40 /* response requested */
#define CMD_FLAGS_VALID 0x80 /* valid descriptor */
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
uint8_t rsp_flags;
uint8_t rsp_numdesc;
uint16_t rsp_id;
uint16_t rsp_seq;
uint16_t rsp_par1;
uint32_t rsp_par2;
uint32_t rsp_par3;
};
#define RSP_FLAGS_TYPE_M 0x07 /* type mask */
#define RSP_FLAGS_TYPE_FRAG 0x00 /* type: fragment */
#define RSP_FLAGS_TYPE_DATA 0x01 /* type: data frame */
#define RSP_FLAGS_TYPE_CMD 0x02 /* type: command frame */
#define RSP_FLAGS_TYPE_OPT 0x03 /* type: options */
#define RSP_FLAGS_TYPE_RX 0x04 /* type: command */
#define RSP_FLAGS_TYPE_RESP 0x05 /* type: response */
#define RSP_FLAGS_ERROR 0x40 /* response error */
struct txp_frag_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
uint8_t frag_flags; /* type/descriptor flags */
uint8_t frag_rsvd1;
uint16_t frag_len; /* bytes in this fragment */
uint32_t frag_addrlo; /* phys addr low word */
uint32_t frag_addrhi; /* phys addr high word */
uint32_t frag_rsvd2;
};
#define FRAG_FLAGS_TYPE_M 0x07 /* type mask */
#define FRAG_FLAGS_TYPE_FRAG 0x00 /* type: fragment */
#define FRAG_FLAGS_TYPE_DATA 0x01 /* type: data frame */
#define FRAG_FLAGS_TYPE_CMD 0x02 /* type: command frame */
#define FRAG_FLAGS_TYPE_OPT 0x03 /* type: options */
#define FRAG_FLAGS_TYPE_RX 0x04 /* type: command */
#define FRAG_FLAGS_TYPE_RESP 0x05 /* type: response */
struct txp_opt_desc {
uint8_t opt_desctype:3,
opt_rsvd:1,
opt_type:4;
uint8_t opt_num;
uint16_t opt_dep1;
uint32_t opt_dep2;
uint32_t opt_dep3;
uint32_t opt_dep4;
};
struct txp_ipsec_desc {
uint8_t ipsec_desctpe:3,
ipsec_rsvd:1,
ipsec_type:4;
uint8_t ipsec_num;
uint16_t ipsec_flags;
uint16_t ipsec_ah1;
uint16_t ipsec_esp1;
uint16_t ipsec_ah2;
uint16_t ipsec_esp2;
uint32_t ipsec_rsvd1;
};
struct txp_tcpseg_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
uint8_t tcpseg_type;
uint8_t tcpseg_num;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new macro TXP_DESC_INC to compute next descriptor index. Tested by: don.nasco <> gmail dot com
2009-03-12 01:14:47 +00:00
uint16_t tcpseg_mss;
uint32_t tcpseg_respaddr;
uint32_t tcpseg_txbytes;
uint32_t tcpseg_lss;
};
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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 TCPSEG_DESC_TYPE_M 0x07 /* type mask */
#define TCPSEG_DESC_TYPE_FRAG 0x00 /* type: fragment */
#define TCPSEG_DESC_TYPE_DATA 0x01 /* type: data frame */
#define TCPSEG_DESC_TYPE_CMD 0x02 /* type: command frame */
#define TCPSEG_DESC_TYPE_OPT 0x03 /* type: options */
#define TCPSEG_DESC_TYPE_RX 0x04 /* type: command */
#define TCPSEG_DESC_TYPE_RESP 0x05 /* type: response */
#define TCPSEG_OPT_IPSEC 0x00
#define TCPSEG_OPT_TSO 0x10
#define TCPSEG_MSS_MASK 0x0FFF
#define TCPSEG_MSS_FIRST 0x1000
#define TCPSEG_MSS_LAST 0x2000
/*
* Transceiver types
*/
#define TXP_XCVR_10_HDX 0
#define TXP_XCVR_10_FDX 1
#define TXP_XCVR_100_HDX 2
#define TXP_XCVR_100_FDX 3
#define TXP_XCVR_AUTO 4
#define TXP_MEDIA_CRC 0x0004 /* crc strip disable */
#define TXP_MEDIA_CD 0x0010 /* collision detection */
#define TXP_MEDIA_CS 0x0020 /* carrier sense */
#define TXP_MEDIA_POL 0x0400 /* polarity reversed */
#define TXP_MEDIA_NOLINK 0x0800 /* 0 = link, 1 = no link */
/*
* receive filter bits (par1 to TXP_CMD_RX_FILTER_{READ|WRITE}
*/
#define TXP_RXFILT_DIRECT 0x0001 /* directed packets */
#define TXP_RXFILT_ALLMULTI 0x0002 /* all multicast packets */
#define TXP_RXFILT_BROADCAST 0x0004 /* broadcast packets */
#define TXP_RXFILT_PROMISC 0x0008 /* promiscuous mode */
#define TXP_RXFILT_HASHMULTI 0x0010 /* use multicast filter */
/*
* boot record (pointers to rings)
*/
struct txp_boot_record {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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 br_hostvar_lo; /* host ring pointer */
uint32_t br_hostvar_hi;
uint32_t br_txlopri_lo; /* tx low pri ring */
uint32_t br_txlopri_hi;
uint32_t br_txlopri_siz;
uint32_t br_txhipri_lo; /* tx high pri ring */
uint32_t br_txhipri_hi;
uint32_t br_txhipri_siz;
uint32_t br_rxlopri_lo; /* rx low pri ring */
uint32_t br_rxlopri_hi;
uint32_t br_rxlopri_siz;
uint32_t br_rxbuf_lo; /* rx buffer ring */
uint32_t br_rxbuf_hi;
uint32_t br_rxbuf_siz;
uint32_t br_cmd_lo; /* command ring */
uint32_t br_cmd_hi;
uint32_t br_cmd_siz;
uint32_t br_resp_lo; /* response ring */
uint32_t br_resp_hi;
uint32_t br_resp_siz;
uint32_t br_zero_lo; /* zero word */
uint32_t br_zero_hi;
uint32_t br_rxhipri_lo; /* rx high pri ring */
uint32_t br_rxhipri_hi;
uint32_t br_rxhipri_siz;
};
/*
* hostvar structure (shared with typhoon)
*/
struct txp_hostvar {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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 hv_rx_hi_read_idx; /* host->arm */
uint32_t hv_rx_lo_read_idx; /* host->arm */
uint32_t hv_rx_buf_write_idx; /* host->arm */
uint32_t hv_resp_read_idx; /* host->arm */
uint32_t hv_tx_lo_desc_read_idx; /* arm->host */
uint32_t hv_tx_hi_desc_read_idx; /* arm->host */
uint32_t hv_rx_lo_write_idx; /* arm->host */
uint32_t hv_rx_buf_read_idx; /* arm->host */
uint32_t hv_cmd_read_idx; /* arm->host */
uint32_t hv_resp_write_idx; /* arm->host */
uint32_t hv_rx_hi_write_idx; /* arm->host */
};
/*
* TYPHOON status register state (in TXP_A2H_0)
*/
#define STAT_ROM_CODE 0x00000001
#define STAT_ROM_EEPROM_LOAD 0x00000002
#define STAT_WAITING_FOR_BOOT 0x00000007
#define STAT_RUNNING 0x00000009
#define STAT_WAITING_FOR_HOST_REQUEST 0x0000000d
#define STAT_WAITING_FOR_SEGMENT 0x00000010
#define STAT_SLEEPING 0x00000011
#define STAT_HALTED 0x00000014
#define TX_ENTRIES 256
#define RX_ENTRIES 128
#define RXBUF_ENTRIES 256
#define CMD_ENTRIES 32
#define RSP_ENTRIES 32
#define OFFLOAD_TCPCKSUM 0x00000002 /* tcp checksum */
#define OFFLOAD_UDPCKSUM 0x00000004 /* udp checksum */
#define OFFLOAD_IPCKSUM 0x00000008 /* ip checksum */
#define OFFLOAD_IPSEC 0x00000010 /* ipsec enable */
#define OFFLOAD_BCAST 0x00000020 /* broadcast throttle */
#define OFFLOAD_DHCP 0x00000040 /* dhcp prevention */
#define OFFLOAD_VLAN 0x00000080 /* vlan enable */
#define OFFLOAD_FILTER 0x00000100 /* filter enable */
#define OFFLOAD_TCPSEG 0x00000200 /* tcp segmentation */
#define OFFLOAD_MASK 0xfffffffe /* mask off low bit */
/*
* Macros for converting array indices to offsets within the descriptor
* arrays. The chip operates on offsets, but it's much easier for us
* to operate on indices. Assumes descriptor entries are 16 bytes.
*/
#define TXP_IDX2OFFSET(idx) ((idx) << 4)
#define TXP_OFFSET2IDX(off) ((off) >> 4)
struct txp_cmd_ring {
struct txp_cmd_desc *base;
uint32_t lastwrite;
uint32_t size;
};
struct txp_rsp_ring {
struct txp_rsp_desc *base;
uint32_t lastwrite;
uint32_t size;
};
struct txp_tx_ring {
struct txp_tx_desc *r_desc; /* base address of descs */
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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 r_tag;
bus_dmamap_t r_map;
uint32_t r_reg; /* register to activate */
uint32_t r_prod; /* producer */
uint32_t r_cons; /* consumer */
uint32_t r_cnt; /* # descs in use */
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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 *r_off; /* hostvar index pointer */
};
struct txp_swdesc {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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 mbuf *sd_mbuf;
bus_dmamap_t sd_map;
};
struct txp_rx_swdesc {
TAILQ_ENTRY(txp_rx_swdesc) sd_next;
struct mbuf *sd_mbuf;
bus_dmamap_t sd_map;
};
struct txp_rx_ring {
struct txp_rx_desc *r_desc; /* base address of descs */
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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 r_tag;
bus_dmamap_t r_map;
uint32_t *r_roff; /* hv read offset ptr */
uint32_t *r_woff; /* hv write offset ptr */
};
struct txp_ldata {
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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_boot_record *txp_boot;
bus_addr_t txp_boot_paddr;
struct txp_hostvar *txp_hostvar;
bus_addr_t txp_hostvar_paddr;
struct txp_tx_desc *txp_txhiring;
bus_addr_t txp_txhiring_paddr;
struct txp_tx_desc *txp_txloring;
bus_addr_t txp_txloring_paddr;
struct txp_rxbuf_desc *txp_rxbufs;
bus_addr_t txp_rxbufs_paddr;
struct txp_rx_desc *txp_rxhiring;
bus_addr_t txp_rxhiring_paddr;
struct txp_rx_desc *txp_rxloring;
bus_addr_t txp_rxloring_paddr;
struct txp_cmd_desc *txp_cmdring;
bus_addr_t txp_cmdring_paddr;
struct txp_rsp_desc *txp_rspring;
bus_addr_t txp_rspring_paddr;
uint32_t *txp_zero;
bus_addr_t txp_zero_paddr;
};
struct txp_chain_data {
bus_dma_tag_t txp_parent_tag;
bus_dma_tag_t txp_boot_tag;
bus_dmamap_t txp_boot_map;
bus_dma_tag_t txp_hostvar_tag;
bus_dmamap_t txp_hostvar_map;
bus_dma_tag_t txp_txhiring_tag;
bus_dmamap_t txp_txhiring_map;
bus_dma_tag_t txp_txloring_tag;
bus_dmamap_t txp_txloring_map;
bus_dma_tag_t txp_tx_tag;
bus_dma_tag_t txp_rx_tag;
bus_dma_tag_t txp_rxbufs_tag;
bus_dmamap_t txp_rxbufs_map;
bus_dma_tag_t txp_rxhiring_tag;
bus_dmamap_t txp_rxhiring_map;
bus_dma_tag_t txp_rxloring_tag;
bus_dmamap_t txp_rxloring_map;
bus_dma_tag_t txp_cmdring_tag;
bus_dmamap_t txp_cmdring_map;
bus_dma_tag_t txp_rspring_tag;
bus_dmamap_t txp_rspring_map;
bus_dma_tag_t txp_zero_tag;
bus_dmamap_t txp_zero_map;
};
struct txp_hw_stats {
uint32_t tx_frames;
uint64_t tx_bytes;
uint32_t tx_deferred;
uint32_t tx_late_colls;
uint32_t tx_colls;
uint32_t tx_carrier_lost;
uint32_t tx_multi_colls;
uint32_t tx_excess_colls;
uint32_t tx_fifo_underruns;
uint32_t tx_mcast_oflows;
uint32_t tx_filtered;
uint32_t rx_frames;
uint64_t rx_bytes;
uint32_t rx_fifo_oflows;
uint32_t rx_badssd;
uint32_t rx_crcerrs;
uint32_t rx_lenerrs;
uint32_t rx_bcast_frames;
uint32_t rx_mcast_frames;
uint32_t rx_oflows;
uint32_t rx_filtered;
};
struct txp_softc {
struct ifnet *sc_ifp;
device_t sc_dev;
struct txp_hostvar *sc_hostvar;
struct txp_boot_record *sc_boot;
struct resource *sc_res;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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 sc_res_id;
int sc_res_type;
struct resource *sc_irq;
void *sc_intrhand;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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_chain_data sc_cdata;
struct txp_ldata sc_ldata;
int sc_rxbufprod;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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 sc_process_limit;
struct txp_cmd_ring sc_cmdring;
struct txp_rsp_ring sc_rspring;
struct callout sc_tick;
struct ifmedia sc_ifmedia;
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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_hw_stats sc_ostats;
struct txp_hw_stats sc_stats;
struct txp_tx_ring sc_txhir, 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
struct txp_swdesc sc_txd[TX_ENTRIES];
struct txp_rxbuf_desc *sc_rxbufs;
struct txp_rx_ring sc_rxhir, sc_rxlor;
uint16_t sc_xcvr;
uint16_t sc_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 sc_watchdog_timer;
int sc_if_flags;
int sc_flags;
#define TXP_FLAG_DETACH 0x4000
#define TXP_FLAG_LINK 0x8000
TAILQ_HEAD(, txp_rx_swdesc) sc_free_list;
TAILQ_HEAD(, txp_rx_swdesc) sc_busy_list;
struct task sc_int_task;
struct taskqueue *sc_tq;
struct mtx sc_mtx;
};
struct txp_fw_file_header {
uint8_t magicid[8]; /* TYPHOON\0 */
uint32_t version;
uint32_t nsections;
uint32_t addr;
uint32_t hmac[5];
};
struct txp_fw_section_header {
uint32_t nbytes;
uint16_t cksum;
uint16_t reserved;
uint32_t addr;
};
#define TXP_MAX_SEGLEN 0xffff
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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_MAX_PKTLEN (ETHER_MAX_LEN + ETHER_VLAN_ENCAP_LEN)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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 WRITE_REG(sc, reg, val) bus_write_4((sc)->sc_res, reg, val)
#define READ_REG(sc, reg) bus_read_4((sc)->sc_res, reg)
#define TXP_BARRIER(sc, o, l, f) bus_barrier((sc)->sc_res, (o), (l), (f))
#define TXP_LOCK(sc) mtx_lock(&(sc)->sc_mtx)
#define TXP_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx)
#define TXP_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED)
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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_MAXTXSEGS 16
#define TXP_RXBUF_ALIGN (sizeof(uint32_t))
#define TXP_PROC_MIN 16
#define TXP_PROC_MAX RX_ENTRIES
#define TXP_PROC_DEFAULT (RX_ENTRIES / 2)
#define TXP_ADDR_HI(x) ((uint64_t)(x) >> 32)
#define TXP_ADDR_LO(x) ((uint64_t)(x) & 0xffffffff)
/*
* 3Com PCI vendor ID.
*/
#define TXP_VENDORID_3COM 0x10B7
/*
* 3cR990 device IDs
*/
#define TXP_DEVICEID_3CR990_TX_95 0x9902
#define TXP_DEVICEID_3CR990_TX_97 0x9903
#define TXP_DEVICEID_3CR990B_TXM 0x9904
#define TXP_DEVICEID_3CR990_SRV_95 0x9908
#define TXP_DEVICEID_3CR990_SRV_97 0x9909
#define TXP_DEVICEID_3CR990B_SRV 0x990A
struct txp_type {
uint16_t txp_vid;
uint16_t txp_did;
char *txp_name;
};
bus_dma(9) conversion and make txp(4) work on all architectures. o Header file cleanup. o bus_dma(9) conversion. - Removed all consumers of vtophys(9) and converted to use bus_dma(9). - Typhoon2 functional specification says the controller supports 64bit DMA addressing. However all Typhoon controllers are known to lack of DAC support so 64bit DMA support was disabled. - The hardware can't handle more 16 fragmented Tx DMA segments so teach txp(4) to collapse these segments to be less than 16. - Added Rx buffer alignment requirements(4 bytes alignment) and implemented fixup code to align receive frame. Previously txp(4) always copied Rx frame to align it on 2 byte boundary but its copy overhead is much higher than unaligned access on i386/amd64. Alignment fixup code is now applied only for strict-alignment architectures. With this change i386 and amd64 will get instant Rx performance boost. Typhoon2 datasheet mentions a command that pads arbitrary bytes in Rx buffer but that command does not work. - Nuked pointer trick in descriptor ring. This does not work on sparc64 and replaced it with bcopy. Alternatively txp(4) can embed a 32 bits index value into the descriptor and compute real buffer address but it may make code complicated. - Added endianness support code in various Tx/Rx/command/response descriptor access. With this change txp(4) should work on all architectures. o Added comments for known firmware bugs(Tx checksum offloading, TSO, VLAN stripping and Rx buffer padding control). o Prefer faster memory space register access to I/O space access. Added fall-back mechanism to use alternative I/O space access. The hardware supports both memory and I/O mapped access. Users can still force to use old I/O space access by setting hw.txp.prefer_iomap tunable to 1 in /boot/loader.conf. o Added experimental suspend/resume methods. o Nuke error prone Rx buffer handling code and implemented local buffer management with TAILQ. Be definition the controller can't pass the last received frame to host if no Rx free buffers are available to use as head and tail pointer of Rx descriptor ring can't have the same value. In that case the Rx buffer pointer in Rx buffer ring still holds a valid buffer and txp_rxbuf_reclaim() can't fill Rx buffers as the first buffer is still valid. Instead of relying on the value of Rx buffer ring, introduce local buffer management code to handle empty buffer situation. This should fix a long standing bug which completely hangs the controller under high network load. I could easily trigger the issue by sending 64 bytes UDP frames with netperf. I have no idea how this bugs was not fixed for a long time. o Converted ithread interrupt handler to filter based one. o Rearranged txp_detach routine such that it's now used for general clean-up routine. o Show sleep image version on device attach time. This will help to know what action should be taken depending on sleep image version. The version information in datasheet was wrong for newer NV images so I followed Linux which seems to correctly extract version numbers from response descriptors. o Firmware image is no longer downloaded in device attach time. Now it is reloaded whenever if_init is invoked. This is to ensure correct operation of hardware when something goes wrong. Previously the controller always run without regard to running state of firmware. This change will add additional controller initialization time but it give more robust operation as txp(4) always start off from a known state. The controller is put into sleep state until administrator explicitly up the interface. o As firmware is loaded in if_init handler, it's now possible to implement real watchdog timeout handler. When watchdog timer is expired, full-reset the controller and initialize the hardware again as most other drivers do. While I'm here use our own timer for watchdog instead of using if_watchdog/if_timer interface. o Instead of masking specific interrupts with TXP_IMR register, program TXP_IER register with the interrupts to be raised and use TXP_IMR to toggle interrupt generation. o Implemented txp_wait() to wait a specific state of a controller. o Separate boot related code from txp_download_fw() and name it txp_boot() to handle boot process. o Added bus_barrier(9) to host to ARM communication. o Added endianness to all typhoon command processing. The ARM93C always expects little-endian format of command/data. o Removed __STRICT_ALIGNMENT which is not valid on FreeBSD. __NO_STRICT_ALIGNMENT is provided for that purpose on FreeBSD. Previously __STRICT_ALIGNMENT was unconditionally defined for all architectures. o Rewrote SIOCSIFCAP ioctl handler such that each capability can be controlled by ifconfig(8). Note, disabling VLAN hardware tagging has no effect due to the bug of firmware. o Don't send TXP_CMD_CLEAR_STATISTICS to clear MAC statistics in txp_tick(). The command is not atomic. Instead, just read the statistics and reflect saved statistics to the statistics. dev.txp.%d.stats sysctl node provides detailed MAC statistics. This also reduces a lot of waste of CPU cycles as processing a command ring takes a very long time on ARM93C. Note, Rx multicast and broadcast statistics does not seem to right. It might be another bug of firmware. o Implemented link state change handling in txp_tick(). Now sending packets is allowed only after establishing a valid link. Also invoke link state change notification whenever its state is changed so pseudo drivers like lagg(4) that relies on link state can work with failover or link aggregation without hacks. if_baudrate is updated to resolved speed so SNMP agents can get correct bandwidth parameters. o Overhauled Tx routine such that it now honors number of allowable DMA segments and checks for 4 free descriptors before trying to send a frame. A frame may require 4 descriptors(1 frame descriptor, 1 or more frame descriptors, 1 TSO option descriptor, one free descriptor to prevent descriptor wrap-around) at least so it's necessary to check available free descriptors prior to setting up DMA operation. o Added a sysctl variable dev.txp.%d.process_limit to control how many received frames should be served in Rx handler. Valid ranges are 16 to 128(default 64) in unit of frames. o Added ALTQ(4) support. o Added missing IFCAP_VLAN_HWCSUM as txp(4) can offload checksum calculation as well as VLAN tag insertion/stripping. o Fixed media header length for VLAN. o Don't set if_mtu in device attach, it's already set in ether_ifattach(). o Enabled MWI. o Fixed module unload panic when bpf listeners are active. o Rearranged ethernet address programming logic such that it works on strict-alignment architectures. o Removed unused member variables in softc. o Added support for WOL. o Removed now unused TXP_PCI_LOMEM/TXP_PCI_LOIO. o Added wakeup command TXP_BOOTCMD_WAKEUP definition. o Added a new firmware version query command, TXP_CMD_READ_VERSION. o Removed volatile keyword in softc as bus_dmamap_sync(9) should take care of this. o Removed embedded union trick of a structure used to to access a pointer on LP64 systems. o Added a few TSO related definitions for struct txp_tcpseg_desc. However TSO is not used at all due to the limitation of hardware. o Redefined PKT_MAX_PKTLEN to theoretical maximum size of a frame. o Switched from bus_space_{read|write}_4 to bus_{read|write}_4. o Added a new 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_TIMEOUT 10000
#define TXP_CMD_NOWAIT 0
#define TXP_CMD_WAIT 1
#define TXP_TX_TIMEOUT 5
/*
* Each frame requires one frame descriptor and one or more
* fragment descriptors. If TSO is used frame descriptor block
* requires one or two option frame descriptors depending on
* number of framents. Therefore we will consume three
* additional descriptors at most to use TSO for a frame and
* one reserved descriptor in order not to full Tx descriptor
* ring.
*/
#define TXP_TXD_RESERVED 4
#define TXP_DESC_INC(x, y) ((x) = ((x) + 1) % (y))