d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
f180142c76
freebsd-nq/sys/dev/gem/if_gem.c

2276 lines
62 KiB
C
Raw Normal View History

Use __FBSDID(). Also some minor style cleanups.
2003-08-24 17:55:58 +00:00
/*-
sys/dev: further adoption of SPDX licensing ID tags. Mainly focus on files that use BSD 2-Clause license, however the tool I was using misidentified many licenses so this was mostly a manual - error prone - task. The Software Package Data Exchange (SPDX) group provides a specification to make it easier for automated tools to detect and summarize well known opensource licenses. We are gradually adopting the specification, noting that the tags are considered only advisory and do not, in any way, superceed or replace the license texts.
2017-11-27 14:52:40 +00:00
* SPDX-License-Identifier: BSD-2-Clause-NetBSD
*
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
* Copyright (C) 2001 Eduardo Horvath.
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
* Copyright (c) 2001-2003 Thomas Moestl
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
* Copyright (c) 2007 Marius Strobl <marius@FreeBSD.org>
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
* 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.
*
* 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 BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
* from: NetBSD: gem.c,v 1.21 2002/06/01 23:50:58 lukem Exp
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
*/
Use __FBSDID(). Also some minor style cleanups.
2003-08-24 17:55:58 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/*
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
* Driver for Apple GMAC, Sun ERI and Sun GEM Ethernet controllers
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
*/
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#if 0
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#define GEM_DEBUG
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Wrap the handler and associated code for collecting completed RX descriptors that are still marked owned in #ifdef GEM_RINT_TIMEOUT instead of #if 0 for convenience. - Remove stale code and comment about relying on the preset XIF config. - In case of a watchdog timeout call the init function instead of just the start function so the chip is properly reset. Merge from hme(4): - Convert to use bus_dmamap_load_mbuf_sg() for loading RX buffers. - Protect from a duplicate mbuf free panic in case the DMA engine hangs. Reviewed by: yongari Tested on: powerpc(grehan), sparc64 MFC after: 1 week
2005-07-24 18:12:31 +00:00
#if 0 /* XXX: In case of emergency, re-enable this. */
#define GEM_RINT_TIMEOUT
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bus.h>
#include <sys/callout.h>
Move the new byte order function prototypes from <sys/param.h> to <sys/endian.h>. This puts us in line with NetBSD and OpenBSD.
2002-04-26 22:48:23 +00:00
#include <sys/endian.h>
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#include <sys/mbuf.h>
#include <sys/malloc.h>
#include <sys/kernel.h>
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
#include <sys/lock.h>
Add missing <sys/module.h> includes currently relying on nested include in <sys/kernel.h>
2004-06-03 06:10:02 +00:00
#include <sys/module.h>
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
#include <sys/mutex.h>
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#include <sys/socket.h>
#include <sys/sockio.h>
Since if_gem is being touted as one of our more architecturally correct network drivers with respect to busmaster DMA, go over it with at duster to make other aspects of it a role model: Eliminate the pci specific softc, it serves no rational purpose. Use convenience resource allocation/deallocation functions to save code and errorhandling. Switch from bus_space_{read|write}_%u() to bus_{read|write}_%u() functions and forget about tags and handles, the resource will know about those, should they be needed. This also eliminates a number of inconsistently named local variables.
2007-05-04 19:15:28 +00:00
#include <sys/rman.h>
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Call bpf_mtap() on outgoing packets.
2002-10-26 12:32:07 +00:00
#include <net/bpf.h>
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#include <net/ethernet.h>
#include <net/if.h>
The r48589 promised to remove implicit inclusion of if_var.h soon. Prepare to this event, adding if_var.h to files that do need it. Also, include all includes that now are included due to implicit pollution via if_var.h Sponsored by: Netflix Sponsored by: Nginx, Inc.
2013-10-26 17:58:36 +00:00
#include <net/if_var.h>
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#include <net/if_arp.h>
#include <net/if_dl.h>
#include <net/if_media.h>
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
#include <net/if_types.h>
- Allow for VLAN-sized frames and set IFCAP_VLAN_MTU. - On resume all registers have to be initialized again like after power-on so reset sc_inited in gem_suspend() in order get all of the registers set next time gem_init_regs() is called. - On at least some ERI and GEM revisions GEM_MAC_RX_OVERFLOW happen often due to a silicon bug and re-initializing is all we can do about these errors so make handling them non-verbose. - Remove a superfluous memset(3) call in gem_meminit(), all elements are initialized to 0 anyway. MFC after: 1 week
2005-08-28 15:07:30 +00:00
#include <net/if_vlan_var.h>
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/ip.h>
#include <netinet/tcp.h>
#include <netinet/udp.h>
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#include <machine/bus.h>
#include <dev/mii/mii.h>
#include <dev/mii/miivar.h>
s=gem/foo=dev/gem/foo=
2003-08-23 06:30:21 +00:00
#include <dev/gem/if_gemreg.h>
#include <dev/gem/if_gemvar.h>
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
CTASSERT(powerof2(GEM_NRXDESC) && GEM_NRXDESC >= 32 && GEM_NRXDESC <= 8192);
CTASSERT(powerof2(GEM_NTXDESC) && GEM_NTXDESC >= 32 && GEM_NTXDESC <= 8192);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
#define GEM_TRIES 10000
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
/*
- Restore the behavior of enabling the MII buffer for an internal PHY only and not also in the case of an external PHY currently doing full duplex, which accidentally got broken in r172334. It's still not clear to me why we need to enable the buffer for an internal PHY though. - Count excess and late collisions as output errors. [1] - Count receive errors as input errors. [1] Obtained from: NetBSD [1] MFC after: 3 days
2008-08-23 15:03:26 +00:00
* The hardware supports basic TCP/UDP checksum offloading. However,
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
* the hardware doesn't compensate the checksum for UDP datagram which
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
* can yield to 0x0. As a safe guard, UDP checksum offload is disabled
* by default. It can be reactivated by setting special link option
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
* link0 with ifconfig(8).
*/
#define GEM_CSUM_FEATURES (CSUM_TCP)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
static int gem_add_rxbuf(struct gem_softc *sc, int idx);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
static int gem_bitwait(struct gem_softc *sc, u_int bank, bus_addr_t r,
uint32_t clr, uint32_t set);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
static void gem_cddma_callback(void *xsc, bus_dma_segment_t *segs,
int nsegs, int error);
static int gem_disable_rx(struct gem_softc *sc);
static int gem_disable_tx(struct gem_softc *sc);
static void gem_eint(struct gem_softc *sc, u_int status);
static void gem_init(void *xsc);
static void gem_init_locked(struct gem_softc *sc);
static void gem_init_regs(struct gem_softc *sc);
static int gem_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
static int gem_load_txmbuf(struct gem_softc *sc, struct mbuf **m_head);
static int gem_meminit(struct gem_softc *sc);
static void gem_mifinit(struct gem_softc *sc);
static void gem_reset(struct gem_softc *sc);
static int gem_reset_rx(struct gem_softc *sc);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
static void gem_reset_rxdma(struct gem_softc *sc);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
static int gem_reset_tx(struct gem_softc *sc);
static u_int gem_ringsize(u_int sz);
static void gem_rint(struct gem_softc *sc);
- Wrap the handler and associated code for collecting completed RX descriptors that are still marked owned in #ifdef GEM_RINT_TIMEOUT instead of #if 0 for convenience. - Remove stale code and comment about relying on the preset XIF config. - In case of a watchdog timeout call the init function instead of just the start function so the chip is properly reset. Merge from hme(4): - Convert to use bus_dmamap_load_mbuf_sg() for loading RX buffers. - Protect from a duplicate mbuf free panic in case the DMA engine hangs. Reviewed by: yongari Tested on: powerpc(grehan), sparc64 MFC after: 1 week
2005-07-24 18:12:31 +00:00
#ifdef GEM_RINT_TIMEOUT
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
static void gem_rint_timeout(void *arg);
Fix warning.
2002-07-24 01:09:38 +00:00
#endif
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
static inline void gem_rxcksum(struct mbuf *m, uint64_t flags);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
static void gem_rxdrain(struct gem_softc *sc);
Correct r223648; as gem_init_locked() was calling gem_setladrf(), which sets GEM_MAC_RX_CONFIG based on sc_mac_rxcfg which in turn is initialized to zero, before reading the supposedly default configuration we were effectively not basing sc_mac_rxcfg and thus GEM_MAC_RX_CONFIG on the default configuration. Solve this by calling gem_setladrf() after reading in the default configuration of GEM_MAC_RX_CONFIG. This also avoids the need to distinguish whether gem_setladrf() should enable the RX MAC again and should be slightly more correct as we're now doing all of the RX MAC configuration in the intended step.
2011-07-12 13:06:44 +00:00
static void gem_setladrf(struct gem_softc *sc);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
static void gem_start(struct ifnet *ifp);
static void gem_start_locked(struct ifnet *ifp);
static void gem_stop(struct ifnet *ifp, int disable);
static void gem_tick(void *arg);
static void gem_tint(struct gem_softc *sc);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
static inline void gem_txkick(struct gem_softc *sc);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
static int gem_watchdog(struct gem_softc *sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
devclass_t gem_devclass;
DRIVER_MODULE(miibus, gem, miibus_driver, miibus_devclass, 0, 0);
MODULE_DEPEND(gem, miibus, 1, 1, 1);
#ifdef GEM_DEBUG
#include <sys/ktr.h>
KTR_CTx are long time aliased by existing classes so they can't serve their purpose anymore. Axe them out. Sponsored by: Sandvine Incorporated Discussed with: jhb, emaste Possible MFC: TBD
2010-07-21 10:05:07 +00:00
#define KTR_GEM KTR_SPARE2
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#endif
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
#define GEM_BANK1_BITWAIT(sc, r, clr, set) \
gem_bitwait((sc), GEM_RES_BANK1, (r), (clr), (set))
#define GEM_BANK2_BITWAIT(sc, r, clr, set) \
gem_bitwait((sc), GEM_RES_BANK2, (r), (clr), (set))
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_attach(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
struct gem_txsoft *txs;
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp;
Convert the PHY drivers to honor the mii_flags passed down and convert the NIC drivers as well as the PHY drivers to take advantage of the mii_attach() introduced in r213878 to get rid of certain hacks. For the most part these were: - Artificially limiting miibus_{read,write}reg methods to certain PHY addresses; we now let mii_attach() only probe the PHY at the desired address(es) instead. - PHY drivers setting MIIF_* flags based on the NIC driver they hang off from, partly even based on grabbing and using the softc of the parent; we now pass these flags down from the NIC to the PHY drivers via mii_attach(). This got us rid of all such hacks except those of brgphy() in combination with bce(4) and bge(4), which is way beyond what can be expressed with simple flags. While at it, I took the opportunity to change the NIC drivers to pass up the error returned by mii_attach() (previously by mii_phy_probe()) and unify the error message used in this case where and as appropriate as mii_attach() actually can fail for a number of reasons, not just because of no PHY(s) being present at the expected address(es). Reviewed by: jhb, yongari
2010-10-15 14:52:11 +00:00
int error, i, phy;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
uint32_t v;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
if (bootverbose)
device_printf(sc->sc_dev, "flags=0x%x\n", sc->sc_flags);
/* Set up ifnet structure. */
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
if (ifp == NULL)
return (ENOSPC);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
sc->sc_csum_features = GEM_CSUM_FEATURES;
ifp->if_softc = sc;
if_initname(ifp, device_get_name(sc->sc_dev),
device_get_unit(sc->sc_dev));
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_start = gem_start;
ifp->if_ioctl = gem_ioctl;
ifp->if_init = gem_init;
IFQ_SET_MAXLEN(&ifp->if_snd, GEM_TXQUEUELEN);
ifp->if_snd.ifq_drv_maxlen = GEM_TXQUEUELEN;
IFQ_SET_READY(&ifp->if_snd);
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
callout_init_mtx(&sc->sc_tick_ch, &sc->sc_mtx, 0);
#ifdef GEM_RINT_TIMEOUT
callout_init_mtx(&sc->sc_rx_ch, &sc->sc_mtx, 0);
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/* Make sure the chip is stopped. */
gem_reset(sc);
add a newbus method for obtaining the bus's bus_dma_tag_t... This is required by arches like sparc64 (not yet implemented) and sun4v where there are seperate IOMMU's for each PCI bus... For all other arches, it will end up returning NULL, which makes it a no-op... Convert a few drivers (the ones we've been working w/ on sun4v) to the new convection... Eventually all drivers will need to replace the parent tag of NULL, w/ bus_get_dma_tag(dev), though dev is usually different for each driver, and will require hand inspection... Reviewed by: scottl (earlier version)
2006-09-03 00:27:42 +00:00
error = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), 1, 0,
BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0, NULL,
NULL, &sc->sc_pdmatag);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
if (error != 0)
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
goto fail_ifnet;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, MCLBYTES,
1, MCLBYTES, BUS_DMA_ALLOCNOW, NULL, NULL, &sc->sc_rdmatag);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
if (error != 0)
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
goto fail_ptag;
error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
MCLBYTES * GEM_NTXSEGS, GEM_NTXSEGS, MCLBYTES,
Mega busdma API commit. Add two new arguments to bus_dma_tag_create(): lockfunc and lockfuncarg. Lockfunc allows a driver to provide a function for managing its locking semantics while using busdma. At the moment, this is used for the asynchronous busdma_swi and callback mechanism. Two lockfunc implementations are provided: busdma_lock_mutex() performs standard mutex operations on the mutex that is specified from lockfuncarg. dftl_lock() is a panic implementation and is defaulted to when NULL, NULL are passed to bus_dma_tag_create(). The only time that NULL, NULL should ever be used is when the driver ensures that bus_dmamap_load() will not be deferred. Drivers that do not provide their own locking can pass busdma_lock_mutex,&Giant args in order to preserve the former behaviour. sparc64 and powerpc do not provide real busdma_swi functions, so this is largely a noop on those platforms. The busdma_swi on is64 is not properly locked yet, so warnings will be emitted on this platform when busdma callback deferrals happen. If anyone gets panics or warnings from dflt_lock() being called, please let me know right away. Reviewed by: tmm, gibbs
2003-07-01 15:52:06 +00:00
BUS_DMA_ALLOCNOW, NULL, NULL, &sc->sc_tdmatag);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
if (error != 0)
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
goto fail_rtag;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
error = bus_dma_tag_create(sc->sc_pdmatag, PAGE_SIZE, 0,
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
sizeof(struct gem_control_data), 1,
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
sizeof(struct gem_control_data), 0,
NULL, NULL, &sc->sc_cdmatag);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
if (error != 0)
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
goto fail_ttag;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/*
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
* Allocate the control data structures, create and load the
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
* DMA map for it.
*/
if ((error = bus_dmamem_alloc(sc->sc_cdmatag,
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
(void **)&sc->sc_control_data,
BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
&sc->sc_cddmamap)) != 0) {
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
device_printf(sc->sc_dev,
"unable to allocate control data, error = %d\n", error);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
goto fail_ctag;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
sc->sc_cddma = 0;
if ((error = bus_dmamap_load(sc->sc_cdmatag, sc->sc_cddmamap,
sc->sc_control_data, sizeof(struct gem_control_data),
gem_cddma_callback, sc, 0)) != 0 || sc->sc_cddma == 0) {
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
device_printf(sc->sc_dev,
"unable to load control data DMA map, error = %d\n",
error);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
goto fail_cmem;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
/*
* Initialize the transmit job descriptors.
*/
STAILQ_INIT(&sc->sc_txfreeq);
STAILQ_INIT(&sc->sc_txdirtyq);
/*
* Create the transmit buffer DMA maps.
*/
error = ENOMEM;
for (i = 0; i < GEM_TXQUEUELEN; i++) {
txs = &sc->sc_txsoft[i];
txs->txs_mbuf = NULL;
txs->txs_ndescs = 0;
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
if ((error = bus_dmamap_create(sc->sc_tdmatag, 0,
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
&txs->txs_dmamap)) != 0) {
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
device_printf(sc->sc_dev,
"unable to create TX DMA map %d, error = %d\n",
i, error);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
goto fail_txd;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
}
/*
* Create the receive buffer DMA maps.
*/
for (i = 0; i < GEM_NRXDESC; i++) {
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
if ((error = bus_dmamap_create(sc->sc_rdmatag, 0,
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
&sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
device_printf(sc->sc_dev,
"unable to create RX DMA map %d, error = %d\n",
i, error);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
goto fail_rxd;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
sc->sc_rxsoft[i].rxs_mbuf = NULL;
}
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
/* Bypass probing PHYs if we already know for sure to use a SERDES. */
if ((sc->sc_flags & GEM_SERDES) != 0)
goto serdes;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
/* Bad things will happen when touching this register on ERI. */
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
if (sc->sc_variant != GEM_SUN_ERI) {
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MII_DATAPATH_MODE,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
GEM_MII_DATAPATH_MII);
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
GEM_BANK1_BARRIER(sc, GEM_MII_DATAPATH_MODE, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
}
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
gem_mifinit(sc);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
/*
* Look for an external PHY.
*/
error = ENXIO;
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
v = GEM_BANK1_READ_4(sc, GEM_MIF_CONFIG);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((v & GEM_MIF_CONFIG_MDI1) != 0) {
v |= GEM_MIF_CONFIG_PHY_SEL;
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MIF_CONFIG, v);
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
GEM_BANK1_BARRIER(sc, GEM_MIF_CONFIG, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
switch (sc->sc_variant) {
case GEM_SUN_ERI:
Convert the PHY drivers to honor the mii_flags passed down and convert the NIC drivers as well as the PHY drivers to take advantage of the mii_attach() introduced in r213878 to get rid of certain hacks. For the most part these were: - Artificially limiting miibus_{read,write}reg methods to certain PHY addresses; we now let mii_attach() only probe the PHY at the desired address(es) instead. - PHY drivers setting MIIF_* flags based on the NIC driver they hang off from, partly even based on grabbing and using the softc of the parent; we now pass these flags down from the NIC to the PHY drivers via mii_attach(). This got us rid of all such hacks except those of brgphy() in combination with bce(4) and bge(4), which is way beyond what can be expressed with simple flags. While at it, I took the opportunity to change the NIC drivers to pass up the error returned by mii_attach() (previously by mii_phy_probe()) and unify the error message used in this case where and as appropriate as mii_attach() actually can fail for a number of reasons, not just because of no PHY(s) being present at the expected address(es). Reviewed by: jhb, yongari
2010-10-15 14:52:11 +00:00
phy = GEM_PHYAD_EXTERNAL;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
break;
default:
Convert the PHY drivers to honor the mii_flags passed down and convert the NIC drivers as well as the PHY drivers to take advantage of the mii_attach() introduced in r213878 to get rid of certain hacks. For the most part these were: - Artificially limiting miibus_{read,write}reg methods to certain PHY addresses; we now let mii_attach() only probe the PHY at the desired address(es) instead. - PHY drivers setting MIIF_* flags based on the NIC driver they hang off from, partly even based on grabbing and using the softc of the parent; we now pass these flags down from the NIC to the PHY drivers via mii_attach(). This got us rid of all such hacks except those of brgphy() in combination with bce(4) and bge(4), which is way beyond what can be expressed with simple flags. While at it, I took the opportunity to change the NIC drivers to pass up the error returned by mii_attach() (previously by mii_phy_probe()) and unify the error message used in this case where and as appropriate as mii_attach() actually can fail for a number of reasons, not just because of no PHY(s) being present at the expected address(es). Reviewed by: jhb, yongari
2010-10-15 14:52:11 +00:00
phy = MII_PHY_ANY;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
break;
}
Convert the PHY drivers to honor the mii_flags passed down and convert the NIC drivers as well as the PHY drivers to take advantage of the mii_attach() introduced in r213878 to get rid of certain hacks. For the most part these were: - Artificially limiting miibus_{read,write}reg methods to certain PHY addresses; we now let mii_attach() only probe the PHY at the desired address(es) instead. - PHY drivers setting MIIF_* flags based on the NIC driver they hang off from, partly even based on grabbing and using the softc of the parent; we now pass these flags down from the NIC to the PHY drivers via mii_attach(). This got us rid of all such hacks except those of brgphy() in combination with bce(4) and bge(4), which is way beyond what can be expressed with simple flags. While at it, I took the opportunity to change the NIC drivers to pass up the error returned by mii_attach() (previously by mii_phy_probe()) and unify the error message used in this case where and as appropriate as mii_attach() actually can fail for a number of reasons, not just because of no PHY(s) being present at the expected address(es). Reviewed by: jhb, yongari
2010-10-15 14:52:11 +00:00
error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
gem_mediachange, gem_mediastatus, BMSR_DEFCAPMASK, phy,
- Fix and enable support for flow control. - Partially revert r172334; as it turns out the DELAYs in gem_reset_{r,t}x() are actually necessary although bus space barriers and gem_bitwait() are used, otherwise the controller may trigger an IOMMU errors on at least sparc64. This is in line with what Linux and OpenSolaris do.
2010-11-22 22:13:26 +00:00
MII_OFFSET_ANY, MIIF_DOPAUSE);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
}
/*
* Fall back on an internal PHY if no external PHY was found.
According to the Linux sungem driver, in case of Apple (K2) GMACs GEM_MIF_CONFIG_MDI0 cannot be trusted when the firmware has powered down the chip so the internal transceiver has to be hardcoded. This is also in line with the AppleGMACEthernet driver, which just doesn't distinguish between internal/external transceiver and MDIO/MDI1 respectively in the first place. Tested by: Andreas Tobler MFC after: 1 week
2010-02-22 21:01:08 +00:00
* Note that with Apple (K2) GMACs GEM_MIF_CONFIG_MDI0 can't be
* trusted when the firmware has powered down the chip.
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
*/
According to the Linux sungem driver, in case of Apple (K2) GMACs GEM_MIF_CONFIG_MDI0 cannot be trusted when the firmware has powered down the chip so the internal transceiver has to be hardcoded. This is also in line with the AppleGMACEthernet driver, which just doesn't distinguish between internal/external transceiver and MDIO/MDI1 respectively in the first place. Tested by: Andreas Tobler MFC after: 1 week
2010-02-22 21:01:08 +00:00
if (error != 0 &&
((v & GEM_MIF_CONFIG_MDI0) != 0 || GEM_IS_APPLE(sc))) {
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
v &= ~GEM_MIF_CONFIG_PHY_SEL;
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MIF_CONFIG, v);
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
GEM_BANK1_BARRIER(sc, GEM_MIF_CONFIG, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
switch (sc->sc_variant) {
case GEM_SUN_ERI:
case GEM_APPLE_K2_GMAC:
Convert the PHY drivers to honor the mii_flags passed down and convert the NIC drivers as well as the PHY drivers to take advantage of the mii_attach() introduced in r213878 to get rid of certain hacks. For the most part these were: - Artificially limiting miibus_{read,write}reg methods to certain PHY addresses; we now let mii_attach() only probe the PHY at the desired address(es) instead. - PHY drivers setting MIIF_* flags based on the NIC driver they hang off from, partly even based on grabbing and using the softc of the parent; we now pass these flags down from the NIC to the PHY drivers via mii_attach(). This got us rid of all such hacks except those of brgphy() in combination with bce(4) and bge(4), which is way beyond what can be expressed with simple flags. While at it, I took the opportunity to change the NIC drivers to pass up the error returned by mii_attach() (previously by mii_phy_probe()) and unify the error message used in this case where and as appropriate as mii_attach() actually can fail for a number of reasons, not just because of no PHY(s) being present at the expected address(es). Reviewed by: jhb, yongari
2010-10-15 14:52:11 +00:00
phy = GEM_PHYAD_INTERNAL;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
break;
case GEM_APPLE_GMAC:
Convert the PHY drivers to honor the mii_flags passed down and convert the NIC drivers as well as the PHY drivers to take advantage of the mii_attach() introduced in r213878 to get rid of certain hacks. For the most part these were: - Artificially limiting miibus_{read,write}reg methods to certain PHY addresses; we now let mii_attach() only probe the PHY at the desired address(es) instead. - PHY drivers setting MIIF_* flags based on the NIC driver they hang off from, partly even based on grabbing and using the softc of the parent; we now pass these flags down from the NIC to the PHY drivers via mii_attach(). This got us rid of all such hacks except those of brgphy() in combination with bce(4) and bge(4), which is way beyond what can be expressed with simple flags. While at it, I took the opportunity to change the NIC drivers to pass up the error returned by mii_attach() (previously by mii_phy_probe()) and unify the error message used in this case where and as appropriate as mii_attach() actually can fail for a number of reasons, not just because of no PHY(s) being present at the expected address(es). Reviewed by: jhb, yongari
2010-10-15 14:52:11 +00:00
phy = GEM_PHYAD_EXTERNAL;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
break;
default:
Convert the PHY drivers to honor the mii_flags passed down and convert the NIC drivers as well as the PHY drivers to take advantage of the mii_attach() introduced in r213878 to get rid of certain hacks. For the most part these were: - Artificially limiting miibus_{read,write}reg methods to certain PHY addresses; we now let mii_attach() only probe the PHY at the desired address(es) instead. - PHY drivers setting MIIF_* flags based on the NIC driver they hang off from, partly even based on grabbing and using the softc of the parent; we now pass these flags down from the NIC to the PHY drivers via mii_attach(). This got us rid of all such hacks except those of brgphy() in combination with bce(4) and bge(4), which is way beyond what can be expressed with simple flags. While at it, I took the opportunity to change the NIC drivers to pass up the error returned by mii_attach() (previously by mii_phy_probe()) and unify the error message used in this case where and as appropriate as mii_attach() actually can fail for a number of reasons, not just because of no PHY(s) being present at the expected address(es). Reviewed by: jhb, yongari
2010-10-15 14:52:11 +00:00
phy = MII_PHY_ANY;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
break;
}
Convert the PHY drivers to honor the mii_flags passed down and convert the NIC drivers as well as the PHY drivers to take advantage of the mii_attach() introduced in r213878 to get rid of certain hacks. For the most part these were: - Artificially limiting miibus_{read,write}reg methods to certain PHY addresses; we now let mii_attach() only probe the PHY at the desired address(es) instead. - PHY drivers setting MIIF_* flags based on the NIC driver they hang off from, partly even based on grabbing and using the softc of the parent; we now pass these flags down from the NIC to the PHY drivers via mii_attach(). This got us rid of all such hacks except those of brgphy() in combination with bce(4) and bge(4), which is way beyond what can be expressed with simple flags. While at it, I took the opportunity to change the NIC drivers to pass up the error returned by mii_attach() (previously by mii_phy_probe()) and unify the error message used in this case where and as appropriate as mii_attach() actually can fail for a number of reasons, not just because of no PHY(s) being present at the expected address(es). Reviewed by: jhb, yongari
2010-10-15 14:52:11 +00:00
error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
gem_mediachange, gem_mediastatus, BMSR_DEFCAPMASK, phy,
- Fix and enable support for flow control. - Partially revert r172334; as it turns out the DELAYs in gem_reset_{r,t}x() are actually necessary although bus space barriers and gem_bitwait() are used, otherwise the controller may trigger an IOMMU errors on at least sparc64. This is in line with what Linux and OpenSolaris do.
2010-11-22 22:13:26 +00:00
MII_OFFSET_ANY, MIIF_DOPAUSE);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
}
/*
* Try the external PCS SERDES if we didn't find any PHYs.
*/
if (error != 0 && sc->sc_variant == GEM_SUN_GEM) {
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
serdes:
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MII_DATAPATH_MODE,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
GEM_MII_DATAPATH_SERDES);
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
GEM_BANK1_BARRIER(sc, GEM_MII_DATAPATH_MODE, 4,
BUS_SPACE_BARRIER_WRITE);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MII_SLINK_CONTROL,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
GEM_MII_SLINK_LOOPBACK | GEM_MII_SLINK_EN_SYNC_D);
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
GEM_BANK1_BARRIER(sc, GEM_MII_SLINK_CONTROL, 4,
BUS_SPACE_BARRIER_WRITE);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MII_CONFIG, GEM_MII_CONFIG_ENABLE);
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
GEM_BANK1_BARRIER(sc, GEM_MII_CONFIG, 4,
BUS_SPACE_BARRIER_WRITE);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
sc->sc_flags |= GEM_SERDES;
Convert the PHY drivers to honor the mii_flags passed down and convert the NIC drivers as well as the PHY drivers to take advantage of the mii_attach() introduced in r213878 to get rid of certain hacks. For the most part these were: - Artificially limiting miibus_{read,write}reg methods to certain PHY addresses; we now let mii_attach() only probe the PHY at the desired address(es) instead. - PHY drivers setting MIIF_* flags based on the NIC driver they hang off from, partly even based on grabbing and using the softc of the parent; we now pass these flags down from the NIC to the PHY drivers via mii_attach(). This got us rid of all such hacks except those of brgphy() in combination with bce(4) and bge(4), which is way beyond what can be expressed with simple flags. While at it, I took the opportunity to change the NIC drivers to pass up the error returned by mii_attach() (previously by mii_phy_probe()) and unify the error message used in this case where and as appropriate as mii_attach() actually can fail for a number of reasons, not just because of no PHY(s) being present at the expected address(es). Reviewed by: jhb, yongari
2010-10-15 14:52:11 +00:00
error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
gem_mediachange, gem_mediastatus, BMSR_DEFCAPMASK,
- Fix and enable support for flow control. - Partially revert r172334; as it turns out the DELAYs in gem_reset_{r,t}x() are actually necessary although bus space barriers and gem_bitwait() are used, otherwise the controller may trigger an IOMMU errors on at least sparc64. This is in line with what Linux and OpenSolaris do.
2010-11-22 22:13:26 +00:00
GEM_PHYAD_EXTERNAL, MII_OFFSET_ANY, MIIF_DOPAUSE);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
}
if (error != 0) {
Convert the PHY drivers to honor the mii_flags passed down and convert the NIC drivers as well as the PHY drivers to take advantage of the mii_attach() introduced in r213878 to get rid of certain hacks. For the most part these were: - Artificially limiting miibus_{read,write}reg methods to certain PHY addresses; we now let mii_attach() only probe the PHY at the desired address(es) instead. - PHY drivers setting MIIF_* flags based on the NIC driver they hang off from, partly even based on grabbing and using the softc of the parent; we now pass these flags down from the NIC to the PHY drivers via mii_attach(). This got us rid of all such hacks except those of brgphy() in combination with bce(4) and bge(4), which is way beyond what can be expressed with simple flags. While at it, I took the opportunity to change the NIC drivers to pass up the error returned by mii_attach() (previously by mii_phy_probe()) and unify the error message used in this case where and as appropriate as mii_attach() actually can fail for a number of reasons, not just because of no PHY(s) being present at the expected address(es). Reviewed by: jhb, yongari
2010-10-15 14:52:11 +00:00
device_printf(sc->sc_dev, "attaching PHYs failed\n");
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
goto fail_rxd;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
sc->sc_mii = device_get_softc(sc->sc_miibus);
/*
* From this point forward, the attachment cannot fail. A failure
* before this point releases all resources that may have been
* allocated.
*/
- Fix some style bugs and remove another banal comment missed in rev. 1.46. - Move the KASSERT on gem_add_rxbuf() to the right spot and add an equivalent one to gem_disable_tx().
2008-03-09 17:55:19 +00:00
/* Get RX FIFO size. */
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
sc->sc_rxfifosize = 64 *
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_READ_4(sc, GEM_RX_FIFO_SIZE);
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
- Fix some style bugs and remove another banal comment missed in rev. 1.46. - Move the KASSERT on gem_add_rxbuf() to the right spot and add an equivalent one to gem_disable_tx().
2008-03-09 17:55:19 +00:00
/* Get TX FIFO size. */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
v = GEM_BANK1_READ_4(sc, GEM_TX_FIFO_SIZE);
Clean up two printf()s that were on a line by themselves unintendedly after the ethernet address printing was moved to common code.
2004-04-23 19:43:35 +00:00
device_printf(sc->sc_dev, "%ukB RX FIFO, %ukB TX FIFO\n",
sc->sc_rxfifosize / 1024, v / 16);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/* Attach the interface. */
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ether_ifattach(ifp, sc->sc_enaddr);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Allow for VLAN-sized frames and set IFCAP_VLAN_MTU. - On resume all registers have to be initialized again like after power-on so reset sc_inited in gem_suspend() in order get all of the registers set next time gem_init_regs() is called. - On at least some ERI and GEM revisions GEM_MAC_RX_OVERFLOW happen often due to a silicon bug and re-initializing is all we can do about these errors so make handling them non-verbose. - Remove a superfluous memset(3) call in gem_meminit(), all elements are initialized to 0 anyway. MFC after: 1 week
2005-08-28 15:07:30 +00:00
/*
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
* Tell the upper layer(s) we support long frames/checksum offloads.
- Allow for VLAN-sized frames and set IFCAP_VLAN_MTU. - On resume all registers have to be initialized again like after power-on so reset sc_inited in gem_suspend() in order get all of the registers set next time gem_init_regs() is called. - On at least some ERI and GEM revisions GEM_MAC_RX_OVERFLOW happen often due to a silicon bug and re-initializing is all we can do about these errors so make handling them non-verbose. - Remove a superfluous memset(3) call in gem_meminit(), all elements are initialized to 0 anyway. MFC after: 1 week
2005-08-28 15:07:30 +00:00
*/
Use define from if_var.h to access a field inside struct if_data, that resides in struct ifnet. Sponsored by: Nginx, Inc.
2014-08-30 19:55:54 +00:00
ifp->if_hdrlen = sizeof(struct ether_vlan_header);
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_HWCSUM;
ifp->if_hwassist |= sc->sc_csum_features;
ifp->if_capenable |= IFCAP_VLAN_MTU | IFCAP_HWCSUM;
- Allow for VLAN-sized frames and set IFCAP_VLAN_MTU. - On resume all registers have to be initialized again like after power-on so reset sc_inited in gem_suspend() in order get all of the registers set next time gem_init_regs() is called. - On at least some ERI and GEM revisions GEM_MAC_RX_OVERFLOW happen often due to a silicon bug and re-initializing is all we can do about these errors so make handling them non-verbose. - Remove a superfluous memset(3) call in gem_meminit(), all elements are initialized to 0 anyway. MFC after: 1 week
2005-08-28 15:07:30 +00:00
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return (0);
/*
* Free any resources we've allocated during the failed attach
* attempt. Do this in reverse order and fall through.
*/
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
fail_rxd:
for (i = 0; i < GEM_NRXDESC; i++)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
bus_dmamap_destroy(sc->sc_rdmatag,
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
sc->sc_rxsoft[i].rxs_dmamap);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
fail_txd:
for (i = 0; i < GEM_TXQUEUELEN; i++)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
if (sc->sc_txsoft[i].txs_dmamap != NULL)
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
bus_dmamap_destroy(sc->sc_tdmatag,
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
sc->sc_txsoft[i].txs_dmamap);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
fail_cmem:
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
sc->sc_cddmamap);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
fail_ctag:
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
bus_dma_tag_destroy(sc->sc_cdmatag);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
fail_ttag:
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
bus_dma_tag_destroy(sc->sc_tdmatag);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
fail_rtag:
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
bus_dma_tag_destroy(sc->sc_rdmatag);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
fail_ptag:
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
bus_dma_tag_destroy(sc->sc_pdmatag);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
fail_ifnet:
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
if_free(ifp);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return (error);
}
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_detach(struct gem_softc *sc)
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
int i;
Revert the part of r194763 which added a dying flag and instead call ether_ifdetach(9) before stopping the controller and the callouts. The consensus is that the latter is now safe to do and should also solve the problem of active BPF listeners clearing promiscuous mode can result in the tick callout being restarted which in turn will trigger a panic once it's actually gone.
2009-06-24 19:04:08 +00:00
ether_ifdetach(ifp);
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_LOCK(sc);
Stop the interface before detaching and freeing it, rather than after. Reported by: marius
2005-06-12 15:33:45 +00:00
gem_stop(ifp, 1);
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_UNLOCK(sc);
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
callout_drain(&sc->sc_tick_ch);
#ifdef GEM_RINT_TIMEOUT
callout_drain(&sc->sc_rx_ch);
#endif
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
if_free(ifp);
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
device_delete_child(sc->sc_dev, sc->sc_miibus);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
for (i = 0; i < GEM_NRXDESC; i++)
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
bus_dmamap_destroy(sc->sc_rdmatag,
sc->sc_rxsoft[i].rxs_dmamap);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
for (i = 0; i < GEM_TXQUEUELEN; i++)
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
if (sc->sc_txsoft[i].txs_dmamap != NULL)
bus_dmamap_destroy(sc->sc_tdmatag,
sc->sc_txsoft[i].txs_dmamap);
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
sc->sc_cddmamap);
bus_dma_tag_destroy(sc->sc_cdmatag);
bus_dma_tag_destroy(sc->sc_tdmatag);
bus_dma_tag_destroy(sc->sc_rdmatag);
bus_dma_tag_destroy(sc->sc_pdmatag);
}
void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_suspend(struct gem_softc *sc)
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_LOCK(sc);
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
gem_stop(ifp, 0);
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_UNLOCK(sc);
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
}
void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_resume(struct gem_softc *sc)
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_LOCK(sc);
- Allow for VLAN-sized frames and set IFCAP_VLAN_MTU. - On resume all registers have to be initialized again like after power-on so reset sc_inited in gem_suspend() in order get all of the registers set next time gem_init_regs() is called. - On at least some ERI and GEM revisions GEM_MAC_RX_OVERFLOW happen often due to a silicon bug and re-initializing is all we can do about these errors so make handling them non-verbose. - Remove a superfluous memset(3) call in gem_meminit(), all elements are initialized to 0 anyway. MFC after: 1 week
2005-08-28 15:07:30 +00:00
/*
* On resume all registers have to be initialized again like
* after power-on.
*/
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
sc->sc_flags &= ~GEM_INITED;
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
if (ifp->if_flags & IFF_UP)
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
gem_init_locked(sc);
GEM_UNLOCK(sc);
Add detach, shutdown, suspend and resume methods. The latter two are not really tested, but are derived from the original NetBSD version.
2003-01-08 20:12:45 +00:00
}
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
static inline void
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
gem_rxcksum(struct mbuf *m, uint64_t flags)
{
struct ether_header *eh;
struct ip *ip;
struct udphdr *uh;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
uint16_t *opts;
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
int32_t hlen, len, pktlen;
uint32_t temp32;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
uint16_t cksum;
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
pktlen = m->m_pkthdr.len;
if (pktlen < sizeof(struct ether_header) + sizeof(struct ip))
return;
eh = mtod(m, struct ether_header *);
if (eh->ether_type != htons(ETHERTYPE_IP))
return;
ip = (struct ip *)(eh + 1);
if (ip->ip_v != IPVERSION)
return;
hlen = ip->ip_hl << 2;
pktlen -= sizeof(struct ether_header);
if (hlen < sizeof(struct ip))
return;
if (ntohs(ip->ip_len) < hlen)
return;
if (ntohs(ip->ip_len) != pktlen)
return;
if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
return; /* Cannot handle fragmented packet. */
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
switch (ip->ip_p) {
case IPPROTO_TCP:
if (pktlen < (hlen + sizeof(struct tcphdr)))
return;
break;
case IPPROTO_UDP:
if (pktlen < (hlen + sizeof(struct udphdr)))
return;
uh = (struct udphdr *)((uint8_t *)ip + hlen);
if (uh->uh_sum == 0)
return; /* no checksum */
break;
default:
return;
}
cksum = ~(flags & GEM_RD_CHECKSUM);
/* checksum fixup for IP options */
len = hlen - sizeof(struct ip);
if (len > 0) {
opts = (uint16_t *)(ip + 1);
for (; len > 0; len -= sizeof(uint16_t), opts++) {
temp32 = cksum - *opts;
temp32 = (temp32 >> 16) + (temp32 & 65535);
cksum = temp32 & 65535;
}
}
m->m_pkthdr.csum_flags |= CSUM_DATA_VALID;
m->m_pkthdr.csum_data = cksum;
}
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_cddma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
struct gem_softc *sc = xsc;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
if (error != 0)
return;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
if (nsegs != 1)
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
panic("%s: bad control buffer segment count", __func__);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
sc->sc_cddma = segs[0].ds_addr;
}
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_tick(void *arg)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
struct gem_softc *sc = arg;
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
struct ifnet *ifp = sc->sc_ifp;
- Restore the behavior of enabling the MII buffer for an internal PHY only and not also in the case of an external PHY currently doing full duplex, which accidentally got broken in r172334. It's still not clear to me why we need to enable the buffer for an internal PHY though. - Count excess and late collisions as output errors. [1] - Count receive errors as input errors. [1] Obtained from: NetBSD [1] MFC after: 3 days
2008-08-23 15:03:26 +00:00
uint32_t v;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
GEM_LOCK_ASSERT(sc, MA_OWNED);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
/*
- Restore the behavior of enabling the MII buffer for an internal PHY only and not also in the case of an external PHY currently doing full duplex, which accidentally got broken in r172334. It's still not clear to me why we need to enable the buffer for an internal PHY though. - Count excess and late collisions as output errors. [1] - Count receive errors as input errors. [1] Obtained from: NetBSD [1] MFC after: 3 days
2008-08-23 15:03:26 +00:00
* Unload collision and error counters.
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
*/
Mechanically convert to if_inc_counter().
2014-09-18 20:21:46 +00:00
if_inc_counter(ifp, IFCOUNTER_COLLISIONS,
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_READ_4(sc, GEM_MAC_NORM_COLL_CNT) +
Mechanically convert to if_inc_counter().
2014-09-18 20:21:46 +00:00
GEM_BANK1_READ_4(sc, GEM_MAC_FIRST_COLL_CNT));
- Restore the behavior of enabling the MII buffer for an internal PHY only and not also in the case of an external PHY currently doing full duplex, which accidentally got broken in r172334. It's still not clear to me why we need to enable the buffer for an internal PHY though. - Count excess and late collisions as output errors. [1] - Count receive errors as input errors. [1] Obtained from: NetBSD [1] MFC after: 3 days
2008-08-23 15:03:26 +00:00
v = GEM_BANK1_READ_4(sc, GEM_MAC_EXCESS_COLL_CNT) +
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_READ_4(sc, GEM_MAC_LATE_COLL_CNT);
Mechanically convert to if_inc_counter().
2014-09-18 20:21:46 +00:00
if_inc_counter(ifp, IFCOUNTER_COLLISIONS, v);
if_inc_counter(ifp, IFCOUNTER_OERRORS, v);
if_inc_counter(ifp, IFCOUNTER_IERRORS,
- Restore the behavior of enabling the MII buffer for an internal PHY only and not also in the case of an external PHY currently doing full duplex, which accidentally got broken in r172334. It's still not clear to me why we need to enable the buffer for an internal PHY though. - Count excess and late collisions as output errors. [1] - Count receive errors as input errors. [1] Obtained from: NetBSD [1] MFC after: 3 days
2008-08-23 15:03:26 +00:00
GEM_BANK1_READ_4(sc, GEM_MAC_RX_LEN_ERR_CNT) +
GEM_BANK1_READ_4(sc, GEM_MAC_RX_ALIGN_ERR) +
GEM_BANK1_READ_4(sc, GEM_MAC_RX_CRC_ERR_CNT) +
Mechanically convert to if_inc_counter().
2014-09-18 20:21:46 +00:00
GEM_BANK1_READ_4(sc, GEM_MAC_RX_CODE_VIOL));
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/*
- Fix some style bugs and remove another banal comment missed in rev. 1.46. - Move the KASSERT on gem_add_rxbuf() to the right spot and add an equivalent one to gem_disable_tx().
2008-03-09 17:55:19 +00:00
* Then clear the hardware counters.
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
*/
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_NORM_COLL_CNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_FIRST_COLL_CNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_EXCESS_COLL_CNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_LATE_COLL_CNT, 0);
- Restore the behavior of enabling the MII buffer for an internal PHY only and not also in the case of an external PHY currently doing full duplex, which accidentally got broken in r172334. It's still not clear to me why we need to enable the buffer for an internal PHY though. - Count excess and late collisions as output errors. [1] - Count receive errors as input errors. [1] Obtained from: NetBSD [1] MFC after: 3 days
2008-08-23 15:03:26 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_LEN_ERR_CNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_ALIGN_ERR, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_CRC_ERR_CNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_CODE_VIOL, 0);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
mii_tick(sc->sc_mii);
- Use the gem_tick() callout instead of if_slowtimo() for driving gem_watchdog() in order to avoid races accessing if_timer. While at it relax the watchdog a bit by reloading it in gem_tint() if there are still packets enqueued. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Fix inconsistencies in prototypes.
2006-12-06 02:04:25 +00:00
if (gem_watchdog(sc) == EJUSTRETURN)
return;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
callout_reset(&sc->sc_tick_ch, hz, gem_tick, sc);
}
static int
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
gem_bitwait(struct gem_softc *sc, u_int bank, bus_addr_t r, uint32_t clr,
uint32_t set)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
int i;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
uint32_t reg;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
for (i = GEM_TRIES; i--; DELAY(100)) {
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
reg = GEM_BANKN_READ_M(bank, 4, sc, r);
- In gem_bitwait() check that the bit clears/was set in the content of the register rather than in the offset describing the register. - In gem_reset_rx() let gem_bitwait() check for the Rx reset bit rather than the Tx reset bit to clear. Obtained from: OpenBSD (same/similar bugs being fixed)
2007-06-16 23:27:59 +00:00
if ((reg & clr) == 0 && (reg & set) == set)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return (1);
}
return (0);
}
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
static void
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
gem_reset(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
CTR2(KTR_GEM, "%s: %s", device_get_name(sc->sc_dev), __func__);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
gem_reset_rx(sc);
gem_reset_tx(sc);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* Do a full reset. */
According to the OpenSolaris source the PCI latency and the cache line size should be set for Sun ERI. Tested by: yongari
2011-07-12 08:20:15 +00:00
GEM_BANK2_WRITE_4(sc, GEM_RESET, GEM_RESET_RX | GEM_RESET_TX |
(sc->sc_variant == GEM_SUN_ERI ? GEM_ERI_CACHE_LINE_SIZE <<
GEM_RESET_CLSZ_SHFT : 0));
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_BANK2_BARRIER(sc, GEM_RESET, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
if (!GEM_BANK2_BITWAIT(sc, GEM_RESET, GEM_RESET_RX | GEM_RESET_TX, 0))
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
device_printf(sc->sc_dev, "cannot reset device\n");
}
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_rxdrain(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
struct gem_rxsoft *rxs;
int i;
for (i = 0; i < GEM_NRXDESC; i++) {
rxs = &sc->sc_rxsoft[i];
if (rxs->rxs_mbuf != NULL) {
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Stop pretending that that we can sync part of a dmamap: replace the GEM_CDTXSYNC and GEM_CDRXSYNC macros with GEM_CDSYNC to sync the complete control map, and combine syncs wherever possible to avoid the overhead. - Sync all maps before unloading them. - Remove a few syncs which should be unnecessary.
2003-01-21 17:17:01 +00:00
bus_dmamap_sync(sc->sc_rdmatag, rxs->rxs_dmamap,
BUS_DMASYNC_POSTREAD);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
bus_dmamap_unload(sc->sc_rdmatag, rxs->rxs_dmamap);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
m_freem(rxs->rxs_mbuf);
rxs->rxs_mbuf = NULL;
}
}
}
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_stop(struct ifnet *ifp, int disable)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
struct gem_softc *sc = ifp->if_softc;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
struct gem_txsoft *txs;
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
CTR2(KTR_GEM, "%s: %s", device_get_name(sc->sc_dev), __func__);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
callout_stop(&sc->sc_tick_ch);
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
#ifdef GEM_RINT_TIMEOUT
callout_stop(&sc->sc_rx_ch);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
gem_reset_tx(sc);
gem_reset_rx(sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/*
* Release any queued transmit buffers.
*/
while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
if (txs->txs_ndescs != 0) {
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Stop pretending that that we can sync part of a dmamap: replace the GEM_CDTXSYNC and GEM_CDRXSYNC macros with GEM_CDSYNC to sync the complete control map, and combine syncs wherever possible to avoid the overhead. - Sync all maps before unloading them. - Remove a few syncs which should be unnecessary.
2003-01-21 17:17:01 +00:00
bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
BUS_DMASYNC_POSTWRITE);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
if (txs->txs_mbuf != NULL) {
m_freem(txs->txs_mbuf);
txs->txs_mbuf = NULL;
}
}
STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
}
if (disable)
gem_rxdrain(sc);
/*
* Mark the interface down and cancel the watchdog timer.
*/
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
sc->sc_flags &= ~GEM_LINK;
- Use the gem_tick() callout instead of if_slowtimo() for driving gem_watchdog() in order to avoid races accessing if_timer. While at it relax the watchdog a bit by reloading it in gem_tint() if there are still packets enqueued. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Fix inconsistencies in prototypes.
2006-12-06 02:04:25 +00:00
sc->sc_wdog_timer = 0;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
static int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_reset_rx(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
/*
* Resetting while DMA is in progress can cause a bus hang, so we
* disable DMA first.
*/
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
(void)gem_disable_rx(sc);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_RX_CONFIG, 0);
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_BANK1_BARRIER(sc, GEM_RX_CONFIG, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
if (!GEM_BANK1_BITWAIT(sc, GEM_RX_CONFIG, GEM_RX_CONFIG_RXDMA_EN, 0))
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
device_printf(sc->sc_dev, "cannot disable RX DMA\n");
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Fix and enable support for flow control. - Partially revert r172334; as it turns out the DELAYs in gem_reset_{r,t}x() are actually necessary although bus space barriers and gem_bitwait() are used, otherwise the controller may trigger an IOMMU errors on at least sparc64. This is in line with what Linux and OpenSolaris do.
2010-11-22 22:13:26 +00:00
/* Wait 5ms extra. */
DELAY(5000);
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
/* Reset the ERX. */
According to the OpenSolaris source the PCI latency and the cache line size should be set for Sun ERI. Tested by: yongari
2011-07-12 08:20:15 +00:00
GEM_BANK2_WRITE_4(sc, GEM_RESET, GEM_RESET_RX |
(sc->sc_variant == GEM_SUN_ERI ? GEM_ERI_CACHE_LINE_SIZE <<
GEM_RESET_CLSZ_SHFT : 0));
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_BANK2_BARRIER(sc, GEM_RESET, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
According to the OpenSolaris source the PCI latency and the cache line size should be set for Sun ERI. Tested by: yongari
2011-07-12 08:20:15 +00:00
if (!GEM_BANK2_BITWAIT(sc, GEM_RESET, GEM_RESET_RX, 0)) {
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
device_printf(sc->sc_dev, "cannot reset receiver\n");
return (1);
}
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
/* Finally, reset RX MAC. */
GEM_BANK1_WRITE_4(sc, GEM_MAC_RXRESET, 1);
GEM_BANK1_BARRIER(sc, GEM_MAC_RXRESET, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
if (!GEM_BANK1_BITWAIT(sc, GEM_MAC_RXRESET, 1, 0)) {
device_printf(sc->sc_dev, "cannot reset RX MAC\n");
return (1);
}
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return (0);
}
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
/*
* Reset the receiver DMA engine.
*
* Intended to be used in case of GEM_INTR_RX_TAG_ERR, GEM_MAC_RX_OVERFLOW
* etc in order to reset the receiver DMA engine only and not do a full
* reset which amongst others also downs the link and clears the FIFOs.
*/
static void
gem_reset_rxdma(struct gem_softc *sc)
{
int i;
Remove unnecessary controller reinitialization by checking IFF_DRV_RUNNING flag. Previously running dhclient or adding alias addresses reinitialized controller and it resulted in unnecessary link flips. Reviewed by: marius
2011-05-20 17:01:22 +00:00
if (gem_reset_rx(sc) != 0) {
sc->sc_ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (gem_init_locked(sc));
Remove unnecessary controller reinitialization by checking IFF_DRV_RUNNING flag. Previously running dhclient or adding alias addresses reinitialized controller and it resulted in unnecessary link flips. Reviewed by: marius
2011-05-20 17:01:22 +00:00
}
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
for (i = 0; i < GEM_NRXDESC; i++)
if (sc->sc_rxsoft[i].rxs_mbuf != NULL)
GEM_UPDATE_RXDESC(sc, i);
sc->sc_rxptr = 0;
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
GEM_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
/* NOTE: we use only 32-bit DMA addresses here. */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_RX_RING_PTR_HI, 0);
GEM_BANK1_WRITE_4(sc, GEM_RX_RING_PTR_LO, GEM_CDRXADDR(sc, 0));
GEM_BANK1_WRITE_4(sc, GEM_RX_KICK, GEM_NRXDESC - 4);
GEM_BANK1_WRITE_4(sc, GEM_RX_CONFIG,
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_ringsize(GEM_NRXDESC /* XXX */) |
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
((ETHER_HDR_LEN + sizeof(struct ip)) <<
GEM_RX_CONFIG_CXM_START_SHFT) |
(GEM_THRSH_1024 << GEM_RX_CONFIG_FIFO_THRS_SHIFT) |
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
(ETHER_ALIGN << GEM_RX_CONFIG_FBOFF_SHFT));
Correct r223648; as gem_init_locked() was calling gem_setladrf(), which sets GEM_MAC_RX_CONFIG based on sc_mac_rxcfg which in turn is initialized to zero, before reading the supposedly default configuration we were effectively not basing sc_mac_rxcfg and thus GEM_MAC_RX_CONFIG on the default configuration. Solve this by calling gem_setladrf() after reading in the default configuration of GEM_MAC_RX_CONFIG. This also avoids the need to distinguish whether gem_setladrf() should enable the RX MAC again and should be slightly more correct as we're now doing all of the RX MAC configuration in the intended step.
2011-07-12 13:06:44 +00:00
/* Adjusting for the SBus clock probably isn't worth the fuzz. */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_RX_BLANKING,
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
((6 * (sc->sc_flags & GEM_PCI66) != 0 ? 2 : 1) <<
GEM_RX_BLANKING_TIME_SHIFT) | 6);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_RX_PAUSE_THRESH,
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
(3 * sc->sc_rxfifosize / 256) |
((sc->sc_rxfifosize / 256) << 12));
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_RX_CONFIG,
GEM_BANK1_READ_4(sc, GEM_RX_CONFIG) | GEM_RX_CONFIG_RXDMA_EN);
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_MASK,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
GEM_MAC_RX_DONE | GEM_MAC_RX_FRAME_CNT);
Correct r223648; as gem_init_locked() was calling gem_setladrf(), which sets GEM_MAC_RX_CONFIG based on sc_mac_rxcfg which in turn is initialized to zero, before reading the supposedly default configuration we were effectively not basing sc_mac_rxcfg and thus GEM_MAC_RX_CONFIG on the default configuration. Solve this by calling gem_setladrf() after reading in the default configuration of GEM_MAC_RX_CONFIG. This also avoids the need to distinguish whether gem_setladrf() should enable the RX MAC again and should be slightly more correct as we're now doing all of the RX MAC configuration in the intended step.
2011-07-12 13:06:44 +00:00
/*
* Clear the RX filter and reprogram it. This will also set the
* current RX MAC configuration and enable it.
*/
gem_setladrf(sc);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
}
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
static int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_reset_tx(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
/*
* Resetting while DMA is in progress can cause a bus hang, so we
* disable DMA first.
*/
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
(void)gem_disable_tx(sc);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_TX_CONFIG, 0);
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_BANK1_BARRIER(sc, GEM_TX_CONFIG, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
if (!GEM_BANK1_BITWAIT(sc, GEM_TX_CONFIG, GEM_TX_CONFIG_TXDMA_EN, 0))
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
device_printf(sc->sc_dev, "cannot disable TX DMA\n");
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Fix and enable support for flow control. - Partially revert r172334; as it turns out the DELAYs in gem_reset_{r,t}x() are actually necessary although bus space barriers and gem_bitwait() are used, otherwise the controller may trigger an IOMMU errors on at least sparc64. This is in line with what Linux and OpenSolaris do.
2010-11-22 22:13:26 +00:00
/* Wait 5ms extra. */
DELAY(5000);
- Fix some style bugs and remove another banal comment missed in rev. 1.46. - Move the KASSERT on gem_add_rxbuf() to the right spot and add an equivalent one to gem_disable_tx().
2008-03-09 17:55:19 +00:00
/* Finally, reset the ETX. */
According to the OpenSolaris source the PCI latency and the cache line size should be set for Sun ERI. Tested by: yongari
2011-07-12 08:20:15 +00:00
GEM_BANK2_WRITE_4(sc, GEM_RESET, GEM_RESET_TX |
(sc->sc_variant == GEM_SUN_ERI ? GEM_ERI_CACHE_LINE_SIZE <<
GEM_RESET_CLSZ_SHFT : 0));
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_BANK2_BARRIER(sc, GEM_RESET, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
According to the OpenSolaris source the PCI latency and the cache line size should be set for Sun ERI. Tested by: yongari
2011-07-12 08:20:15 +00:00
if (!GEM_BANK2_BITWAIT(sc, GEM_RESET, GEM_RESET_TX, 0)) {
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
device_printf(sc->sc_dev, "cannot reset transmitter\n");
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return (1);
}
return (0);
}
static int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_disable_rx(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_CONFIG,
GEM_BANK1_READ_4(sc, GEM_MAC_RX_CONFIG) & ~GEM_MAC_RX_ENABLE);
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_BANK1_BARRIER(sc, GEM_MAC_RX_CONFIG, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
if (GEM_BANK1_BITWAIT(sc, GEM_MAC_RX_CONFIG, GEM_MAC_RX_ENABLE, 0))
return (1);
device_printf(sc->sc_dev, "cannot disable RX MAC\n");
return (0);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
static int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_disable_tx(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_TX_CONFIG,
GEM_BANK1_READ_4(sc, GEM_MAC_TX_CONFIG) & ~GEM_MAC_TX_ENABLE);
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_BANK1_BARRIER(sc, GEM_MAC_TX_CONFIG, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
if (GEM_BANK1_BITWAIT(sc, GEM_MAC_TX_CONFIG, GEM_MAC_TX_ENABLE, 0))
return (1);
device_printf(sc->sc_dev, "cannot disable TX MAC\n");
return (0);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
static int
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
gem_meminit(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
struct gem_rxsoft *rxs;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
int error, i;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
GEM_LOCK_ASSERT(sc, MA_OWNED);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/*
* Initialize the transmit descriptor ring.
*/
for (i = 0; i < GEM_NTXDESC; i++) {
sc->sc_txdescs[i].gd_flags = 0;
sc->sc_txdescs[i].gd_addr = 0;
}
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
sc->sc_txfree = GEM_MAXTXFREE;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
sc->sc_txnext = 0;
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
sc->sc_txwin = 0;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/*
* Initialize the receive descriptor and receive job
* descriptor rings.
*/
for (i = 0; i < GEM_NRXDESC; i++) {
rxs = &sc->sc_rxsoft[i];
if (rxs->rxs_mbuf == NULL) {
if ((error = gem_add_rxbuf(sc, i)) != 0) {
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
device_printf(sc->sc_dev,
"unable to allocate or map RX buffer %d, "
"error = %d\n", i, error);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/*
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
* XXX we should attempt to run with fewer
* receive buffers instead of just failing.
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
*/
gem_rxdrain(sc);
return (1);
}
} else
GEM_INIT_RXDESC(sc, i);
}
sc->sc_rxptr = 0;
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
GEM_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return (0);
}
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
static u_int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_ringsize(u_int sz)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
switch (sz) {
case 32:
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (GEM_RING_SZ_32);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
case 64:
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (GEM_RING_SZ_64);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
case 128:
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (GEM_RING_SZ_128);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
case 256:
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (GEM_RING_SZ_256);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
case 512:
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (GEM_RING_SZ_512);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
case 1024:
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (GEM_RING_SZ_1024);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
case 2048:
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (GEM_RING_SZ_2048);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
case 4096:
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (GEM_RING_SZ_4096);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
case 8192:
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (GEM_RING_SZ_8192);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
default:
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
printf("%s: invalid ring size %d\n", __func__, sz);
return (GEM_RING_SZ_32);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
}
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_init(void *xsc)
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
{
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
struct gem_softc *sc = xsc;
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_LOCK(sc);
gem_init_locked(sc);
GEM_UNLOCK(sc);
}
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/*
* Initialization of interface; set up initialization block
* and transmit/receive descriptor rings.
*/
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_init_locked(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
uint32_t v;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_LOCK_ASSERT(sc, MA_OWNED);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Remove unnecessary controller reinitialization by checking IFF_DRV_RUNNING flag. Previously running dhclient or adding alias addresses reinitialized controller and it resulted in unnecessary link flips. Reviewed by: marius
2011-05-20 17:01:22 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
return;
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
CTR2(KTR_GEM, "%s: %s: calling stop", device_get_name(sc->sc_dev),
__func__);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/*
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
* Initialization sequence. The numbered steps below correspond
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
* to the sequence outlined in section 6.3.5.1 in the Ethernet
* Channel Engine manual (part of the PCIO manual).
* See also the STP2002-STQ document from Sun Microsystems.
*/
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* step 1 & 2. Reset the Ethernet Channel. */
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
gem_stop(ifp, 0);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
gem_reset(sc);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
CTR2(KTR_GEM, "%s: %s: restarting", device_get_name(sc->sc_dev),
__func__);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
if ((sc->sc_flags & GEM_SERDES) == 0)
/* Re-initialize the MIF. */
gem_mifinit(sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* step 3. Setup data structures in host memory. */
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if (gem_meminit(sc) != 0)
return;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* step 4. TX MAC registers & counters */
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
gem_init_regs(sc);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* step 5. RX MAC registers & counters */
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* step 6 & 7. Program Descriptor Ring Base Addresses. */
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/* NOTE: we use only 32-bit DMA addresses here. */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_TX_RING_PTR_HI, 0);
GEM_BANK1_WRITE_4(sc, GEM_TX_RING_PTR_LO, GEM_CDTXADDR(sc, 0));
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_RX_RING_PTR_HI, 0);
GEM_BANK1_WRITE_4(sc, GEM_RX_RING_PTR_LO, GEM_CDRXADDR(sc, 0));
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
CTR3(KTR_GEM, "loading RX ring %lx, TX ring %lx, cddma %lx",
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
GEM_CDRXADDR(sc, 0), GEM_CDTXADDR(sc, 0), sc->sc_cddma);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* step 8. Global Configuration & Interrupt Mask */
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
/*
* Set the internal arbitration to "infinite" bursts of the
* maximum length of 31 * 64 bytes so DMA transfers aren't
* split up in cache line size chunks. This greatly improves
* RX performance.
* Enable silicon bug workarounds for the Apple variants.
*/
GEM_BANK1_WRITE_4(sc, GEM_CONFIG,
GEM_CONFIG_TXDMA_LIMIT | GEM_CONFIG_RXDMA_LIMIT |
((sc->sc_flags & GEM_PCI) != 0 ? GEM_CONFIG_BURST_INF :
GEM_CONFIG_BURST_64) | (GEM_IS_APPLE(sc) ?
GEM_CONFIG_RONPAULBIT | GEM_CONFIG_BUG2FIX : 0));
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_INTMASK,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
~(GEM_INTR_TX_INTME | GEM_INTR_TX_EMPTY | GEM_INTR_RX_DONE |
GEM_INTR_RX_NOBUF | GEM_INTR_RX_TAG_ERR | GEM_INTR_PERR |
GEM_INTR_BERR
#ifdef GEM_DEBUG
| GEM_INTR_PCS | GEM_INTR_MIF
#endif
));
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_MASK,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
GEM_MAC_RX_DONE | GEM_MAC_RX_FRAME_CNT);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_TX_MASK,
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
GEM_MAC_TX_XMIT_DONE | GEM_MAC_TX_DEFER_EXP |
GEM_MAC_TX_PEAK_EXP);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
#ifdef GEM_DEBUG
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_CONTROL_MASK,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
~(GEM_MAC_PAUSED | GEM_MAC_PAUSE | GEM_MAC_RESUME));
#else
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_CONTROL_MASK,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
GEM_MAC_PAUSED | GEM_MAC_PAUSE | GEM_MAC_RESUME);
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* step 9. ETX Configuration: use mostly default values. */
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* Enable DMA. */
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
v = gem_ringsize(GEM_NTXDESC);
/* Set TX FIFO threshold and enable DMA. */
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
v |= ((sc->sc_variant == GEM_SUN_ERI ? 0x100 : 0x4ff) << 10) &
GEM_TX_CONFIG_TXFIFO_TH;
GEM_BANK1_WRITE_4(sc, GEM_TX_CONFIG, v | GEM_TX_CONFIG_TXDMA_EN);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* step 10. ERX Configuration */
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
/* Encode Receive Descriptor ring size. */
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
v = gem_ringsize(GEM_NRXDESC /* XXX */);
/* RX TCP/UDP checksum offset */
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
v |= ((ETHER_HDR_LEN + sizeof(struct ip)) <<
GEM_RX_CONFIG_CXM_START_SHFT);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
/* Set RX FIFO threshold, set first byte offset and enable DMA. */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_RX_CONFIG,
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
v | (GEM_THRSH_1024 << GEM_RX_CONFIG_FIFO_THRS_SHIFT) |
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
(ETHER_ALIGN << GEM_RX_CONFIG_FBOFF_SHFT) |
GEM_RX_CONFIG_RXDMA_EN);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
Correct r223648; as gem_init_locked() was calling gem_setladrf(), which sets GEM_MAC_RX_CONFIG based on sc_mac_rxcfg which in turn is initialized to zero, before reading the supposedly default configuration we were effectively not basing sc_mac_rxcfg and thus GEM_MAC_RX_CONFIG on the default configuration. Solve this by calling gem_setladrf() after reading in the default configuration of GEM_MAC_RX_CONFIG. This also avoids the need to distinguish whether gem_setladrf() should enable the RX MAC again and should be slightly more correct as we're now doing all of the RX MAC configuration in the intended step.
2011-07-12 13:06:44 +00:00
/* Adjusting for the SBus clock probably isn't worth the fuzz. */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_RX_BLANKING,
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
((6 * (sc->sc_flags & GEM_PCI66) != 0 ? 2 : 1) <<
GEM_RX_BLANKING_TIME_SHIFT) | 6);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/*
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
* The following value is for an OFF Threshold of about 3/4 full
* and an ON Threshold of 1/4 full.
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
*/
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_RX_PAUSE_THRESH,
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
(3 * sc->sc_rxfifosize / 256) |
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
((sc->sc_rxfifosize / 256) << 12));
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* step 11. Configure Media. */
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* step 12. RX_MAC Configuration Register */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
v = GEM_BANK1_READ_4(sc, GEM_MAC_RX_CONFIG);
Correct r223648; as gem_init_locked() was calling gem_setladrf(), which sets GEM_MAC_RX_CONFIG based on sc_mac_rxcfg which in turn is initialized to zero, before reading the supposedly default configuration we were effectively not basing sc_mac_rxcfg and thus GEM_MAC_RX_CONFIG on the default configuration. Solve this by calling gem_setladrf() after reading in the default configuration of GEM_MAC_RX_CONFIG. This also avoids the need to distinguish whether gem_setladrf() should enable the RX MAC again and should be slightly more correct as we're now doing all of the RX MAC configuration in the intended step.
2011-07-12 13:06:44 +00:00
v &= ~GEM_MAC_RX_ENABLE;
v |= GEM_MAC_RX_STRIP_CRC;
sc->sc_mac_rxcfg = v;
/*
* Clear the RX filter and reprogram it. This will also set the
* current RX MAC configuration and enable it.
*/
gem_setladrf(sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
/* step 13. TX_MAC Configuration Register */
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
v = GEM_BANK1_READ_4(sc, GEM_MAC_TX_CONFIG);
v |= GEM_MAC_TX_ENABLE;
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
(void)gem_disable_tx(sc);
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_TX_CONFIG, v);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* step 14. Issue Transmit Pending command. */
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Correct spelling in comments. Submitted by: brucec
2011-04-22 09:41:51 +00:00
/* step 15. Give the receiver a swift kick. */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_RX_KICK, GEM_NRXDESC - 4);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
ifp->if_drv_flags |= IFF_DRV_RUNNING;
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
mii_mediachg(sc->sc_mii);
/* Start the one second timer. */
sc->sc_wdog_timer = 0;
callout_reset(&sc->sc_tick_ch, hz, gem_tick, sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
static int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_load_txmbuf(struct gem_softc *sc, struct mbuf **m_head)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
bus_dma_segment_t txsegs[GEM_NTXSEGS];
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
struct gem_txsoft *txs;
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
struct ip *ip;
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
struct mbuf *m;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
uint64_t cflags, flags;
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
int error, nexttx, nsegs, offset, seg;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
GEM_LOCK_ASSERT(sc, MA_OWNED);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
/* Get a work queue entry. */
if ((txs = STAILQ_FIRST(&sc->sc_txfreeq)) == NULL) {
/* Ran out of descriptors. */
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
return (ENOBUFS);
}
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
cflags = 0;
if (((*m_head)->m_pkthdr.csum_flags & sc->sc_csum_features) != 0) {
if (M_WRITABLE(*m_head) == 0) {
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m = m_dup(*m_head, M_NOWAIT);
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
m_freem(*m_head);
*m_head = m;
if (m == NULL)
return (ENOBUFS);
}
offset = sizeof(struct ether_header);
m = m_pullup(*m_head, offset + sizeof(struct ip));
if (m == NULL) {
*m_head = NULL;
return (ENOBUFS);
}
ip = (struct ip *)(mtod(m, caddr_t) + offset);
offset += (ip->ip_hl << 2);
cflags = offset << GEM_TD_CXSUM_STARTSHFT |
((offset + m->m_pkthdr.csum_data) <<
GEM_TD_CXSUM_STUFFSHFT) | GEM_TD_CXSUM_ENABLE;
*m_head = m;
}
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag, txs->txs_dmamap,
*m_head, txsegs, &nsegs, BUS_DMA_NOWAIT);
if (error == EFBIG) {
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m = m_collapse(*m_head, M_NOWAIT, GEM_NTXSEGS);
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
if (m == NULL) {
m_freem(*m_head);
*m_head = NULL;
return (ENOBUFS);
}
*m_head = m;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag,
txs->txs_dmamap, *m_head, txsegs, &nsegs,
BUS_DMA_NOWAIT);
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
if (error != 0) {
m_freem(*m_head);
*m_head = NULL;
return (error);
}
} else if (error != 0)
return (error);
- Fix some style bugs and remove another banal comment missed in rev. 1.46. - Move the KASSERT on gem_add_rxbuf() to the right spot and add an equivalent one to gem_disable_tx().
2008-03-09 17:55:19 +00:00
/* If nsegs is wrong then the stack is corrupt. */
KASSERT(nsegs <= GEM_NTXSEGS,
("%s: too many DMA segments (%d)", __func__, nsegs));
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
if (nsegs == 0) {
m_freem(*m_head);
*m_head = NULL;
return (EIO);
}
/*
* Ensure we have enough descriptors free to describe
* the packet. Note, we always reserve one descriptor
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
* at the end of the ring as a termination point, in
* order to prevent wrap-around.
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
*/
if (nsegs > sc->sc_txfree - 1) {
txs->txs_ndescs = 0;
bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
return (ENOBUFS);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
}
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
txs->txs_ndescs = nsegs;
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
txs->txs_firstdesc = sc->sc_txnext;
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
nexttx = txs->txs_firstdesc;
for (seg = 0; seg < nsegs; seg++, nexttx = GEM_NEXTTX(nexttx)) {
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
#ifdef GEM_DEBUG
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
CTR6(KTR_GEM,
"%s: mapping seg %d (txd %d), len %lx, addr %#lx (%#lx)",
__func__, seg, nexttx, txsegs[seg].ds_len,
txsegs[seg].ds_addr,
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
GEM_DMA_WRITE(sc, txsegs[seg].ds_addr));
#endif
sc->sc_txdescs[nexttx].gd_addr =
GEM_DMA_WRITE(sc, txsegs[seg].ds_addr);
KASSERT(txsegs[seg].ds_len < GEM_TD_BUFSIZE,
("%s: segment size too large!", __func__));
flags = txsegs[seg].ds_len & GEM_TD_BUFSIZE;
sc->sc_txdescs[nexttx].gd_flags =
GEM_DMA_WRITE(sc, flags | cflags);
txs->txs_lastdesc = nexttx;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* Set EOP on the last descriptor. */
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
#ifdef GEM_DEBUG
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
CTR3(KTR_GEM, "%s: end of packet at segment %d, TX %d",
__func__, seg, nexttx);
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
#endif
sc->sc_txdescs[txs->txs_lastdesc].gd_flags |=
GEM_DMA_WRITE(sc, GEM_TD_END_OF_PACKET);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* Lastly set SOP on the first descriptor. */
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
#ifdef GEM_DEBUG
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
CTR3(KTR_GEM, "%s: start of packet at segment %d, TX %d",
__func__, seg, nexttx);
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
#endif
if (++sc->sc_txwin > GEM_NTXSEGS * 2 / 3) {
sc->sc_txwin = 0;
sc->sc_txdescs[txs->txs_firstdesc].gd_flags |=
GEM_DMA_WRITE(sc, GEM_TD_INTERRUPT_ME |
GEM_TD_START_OF_PACKET);
} else
sc->sc_txdescs[txs->txs_firstdesc].gd_flags |=
GEM_DMA_WRITE(sc, GEM_TD_START_OF_PACKET);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
/* Sync the DMA map. */
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
BUS_DMASYNC_PREWRITE);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
CTR4(KTR_GEM, "%s: setting firstdesc=%d, lastdesc=%d, ndescs=%d",
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
__func__, txs->txs_firstdesc, txs->txs_lastdesc,
txs->txs_ndescs);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
STAILQ_REMOVE_HEAD(&sc->sc_txfreeq, txs_q);
STAILQ_INSERT_TAIL(&sc->sc_txdirtyq, txs, txs_q);
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
txs->txs_mbuf = *m_head;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
sc->sc_txnext = GEM_NEXTTX(txs->txs_lastdesc);
sc->sc_txfree -= txs->txs_ndescs;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
return (0);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_init_regs(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
- Store pointer to the link-level address right in "struct ifnet" rather than in ifindex_table[]; all (except one) accesses are through ifp anyway. IF_LLADDR() works faster, and all (except one) ifaddr_byindex() users were converted to use ifp->if_addr. - Stop storing a (pointer to) Ethernet address in "struct arpcom", and drop the IFP2ENADDR() macro; all users have been converted to use IF_LLADDR() instead.
2005-11-11 16:04:59 +00:00
const u_char *laddr = IF_LLADDR(sc->sc_ifp);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
GEM_LOCK_ASSERT(sc, MA_OWNED);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* These registers are not cleared on reset. */
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((sc->sc_flags & GEM_INITED) == 0) {
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* magic values */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_IPG0, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_IPG1, 8);
GEM_BANK1_WRITE_4(sc, GEM_MAC_IPG2, 4);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
/* min frame length */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_MAC_MIN_FRAME, ETHER_MIN_LEN);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
/* max frame length and max burst size */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_MAC_MAX_FRAME,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
(ETHER_MAX_LEN + ETHER_VLAN_ENCAP_LEN) | (0x2000 << 16));
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
/* more magic values */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_PREAMBLE_LEN, 0x7);
GEM_BANK1_WRITE_4(sc, GEM_MAC_JAM_SIZE, 0x4);
GEM_BANK1_WRITE_4(sc, GEM_MAC_ATTEMPT_LIMIT, 0x10);
- Fix and enable support for flow control. - Partially revert r172334; as it turns out the DELAYs in gem_reset_{r,t}x() are actually necessary although bus space barriers and gem_bitwait() are used, otherwise the controller may trigger an IOMMU errors on at least sparc64. This is in line with what Linux and OpenSolaris do.
2010-11-22 22:13:26 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_CONTROL_TYPE, 0x8808);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
/* random number seed */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_RANDOM_SEED,
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
((laddr[5] << 8) | laddr[4]) & 0x3ff);
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* secondary MAC address: 0:0:0:0:0:0 */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADDR3, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADDR4, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADDR5, 0);
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* MAC control address: 01:80:c2:00:00:01 */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADDR6, 0x0001);
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADDR7, 0xc200);
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADDR8, 0x0180);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* MAC filter address: 0:0:0:0:0:0 */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADDR_FILTER0, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADDR_FILTER1, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADDR_FILTER2, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADR_FLT_MASK1_2, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADR_FLT_MASK0, 0);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
sc->sc_flags |= GEM_INITED;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* Counters need to be zeroed. */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_NORM_COLL_CNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_FIRST_COLL_CNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_EXCESS_COLL_CNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_LATE_COLL_CNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_DEFER_TMR_CNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_PEAK_ATTEMPTS, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_FRAME_COUNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_LEN_ERR_CNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_ALIGN_ERR, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_CRC_ERR_CNT, 0);
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_CODE_VIOL, 0);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
/* Set XOFF PAUSE time. */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_SEND_PAUSE_CMD, 0x1BF0);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* Set the station address. */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADDR0, (laddr[4] << 8) | laddr[5]);
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADDR1, (laddr[2] << 8) | laddr[3]);
GEM_BANK1_WRITE_4(sc, GEM_MAC_ADDR2, (laddr[0] << 8) | laddr[1]);
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
/* Enable MII outputs. */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_XIF_CONFIG, GEM_MAC_XIF_TX_MII_ENA);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_start(struct ifnet *ifp)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
struct gem_softc *sc = ifp->if_softc;
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_LOCK(sc);
gem_start_locked(ifp);
GEM_UNLOCK(sc);
}
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
static inline void
gem_txkick(struct gem_softc *sc)
{
/*
* Update the TX kick register. This register has to point to the
* descriptor after the last valid one and for optimum performance
* should be incremented in multiples of 4 (the DMA engine fetches/
* updates descriptors in batches of 4).
*/
#ifdef GEM_DEBUG
CTR3(KTR_GEM, "%s: %s: kicking TX %d",
device_get_name(sc->sc_dev), __func__, sc->sc_txnext);
#endif
GEM_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
GEM_BANK1_WRITE_4(sc, GEM_TX_KICK, sc->sc_txnext);
}
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_start_locked(struct ifnet *ifp)
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
{
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
struct gem_softc *sc = ifp->if_softc;
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
struct mbuf *m;
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
int kicked, ntx;
GEM_LOCK_ASSERT(sc, MA_OWNED);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
IFF_DRV_RUNNING || (sc->sc_flags & GEM_LINK) == 0)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return;
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
CTR4(KTR_GEM, "%s: %s: txfree %d, txnext %d",
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
device_get_name(sc->sc_dev), __func__, sc->sc_txfree,
sc->sc_txnext);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
ntx = 0;
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
kicked = 0;
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
for (; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) && sc->sc_txfree > 1;) {
IFQ_DRV_DEQUEUE(&ifp->if_snd, m);
if (m == NULL)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
break;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if (gem_load_txmbuf(sc, &m) != 0) {
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
if (m == NULL)
break;
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
IFQ_DRV_PREPEND(&ifp->if_snd, m);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
break;
}
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
if ((sc->sc_txnext % 4) == 0) {
gem_txkick(sc);
kicked = 1;
} else
kicked = 0;
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
ntx++;
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
BPF_MTAP(ifp, m);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
if (ntx > 0) {
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
if (kicked == 0)
gem_txkick(sc);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
CTR2(KTR_GEM, "%s: packets enqueued, OWN on %d",
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
device_get_name(sc->sc_dev), sc->sc_txnext);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/* Set a watchdog timer in case the chip flakes out. */
- Use the gem_tick() callout instead of if_slowtimo() for driving gem_watchdog() in order to avoid races accessing if_timer. While at it relax the watchdog a bit by reloading it in gem_tint() if there are still packets enqueued. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Fix inconsistencies in prototypes.
2006-12-06 02:04:25 +00:00
sc->sc_wdog_timer = 5;
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
CTR3(KTR_GEM, "%s: %s: watchdog %d",
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
device_get_name(sc->sc_dev), __func__,
sc->sc_wdog_timer);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
}
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_tint(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
struct gem_txsoft *txs;
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
int progress;
uint32_t txlast;
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
int i;
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
GEM_LOCK_ASSERT(sc, MA_OWNED);
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
CTR2(KTR_GEM, "%s: %s", device_get_name(sc->sc_dev), __func__);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/*
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
* Go through our TX list and free mbufs for those
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
* frames that have been transmitted.
*/
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
progress = 0;
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Stop pretending that that we can sync part of a dmamap: replace the GEM_CDTXSYNC and GEM_CDRXSYNC macros with GEM_CDSYNC to sync the complete control map, and combine syncs wherever possible to avoid the overhead. - Sync all maps before unloading them. - Remove a few syncs which should be unnecessary.
2003-01-21 17:17:01 +00:00
GEM_CDSYNC(sc, BUS_DMASYNC_POSTREAD);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
#ifdef GEM_DEBUG
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
if ((ifp->if_flags & IFF_DEBUG) != 0) {
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
printf(" txsoft %p transmit chain:\n", txs);
for (i = txs->txs_firstdesc;; i = GEM_NEXTTX(i)) {
printf("descriptor %d: ", i);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
printf("gd_flags: 0x%016llx\t",
(long long)GEM_DMA_READ(sc,
sc->sc_txdescs[i].gd_flags));
printf("gd_addr: 0x%016llx\n",
(long long)GEM_DMA_READ(sc,
sc->sc_txdescs[i].gd_addr));
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
if (i == txs->txs_lastdesc)
break;
}
}
#endif
/*
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
* In theory, we could harvest some descriptors before
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
* the ring is empty, but that's a bit complicated.
*
* GEM_TX_COMPLETION points to the last descriptor
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
* processed + 1.
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
*/
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
txlast = GEM_BANK1_READ_4(sc, GEM_TX_COMPLETION);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
CTR4(KTR_GEM, "%s: txs->txs_firstdesc = %d, "
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
"txs->txs_lastdesc = %d, txlast = %d",
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
__func__, txs->txs_firstdesc, txs->txs_lastdesc, txlast);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
if (txs->txs_firstdesc <= txs->txs_lastdesc) {
if ((txlast >= txs->txs_firstdesc) &&
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
(txlast <= txs->txs_lastdesc))
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
break;
} else {
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* Ick -- this command wraps. */
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
if ((txlast >= txs->txs_firstdesc) ||
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
(txlast <= txs->txs_lastdesc))
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
break;
}
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
CTR1(KTR_GEM, "%s: releasing a descriptor", __func__);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
sc->sc_txfree += txs->txs_ndescs;
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
BUS_DMASYNC_POSTWRITE);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
if (txs->txs_mbuf != NULL) {
m_freem(txs->txs_mbuf);
txs->txs_mbuf = NULL;
}
STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
Mechanically convert to if_inc_counter().
2014-09-18 20:21:46 +00:00
if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
progress = 1;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
CTR4(KTR_GEM, "%s: GEM_TX_STATE_MACHINE %x GEM_TX_DATA_PTR %llx "
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
"GEM_TX_COMPLETION %x",
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
__func__, GEM_BANK1_READ_4(sc, GEM_TX_STATE_MACHINE),
((long long)GEM_BANK1_READ_4(sc, GEM_TX_DATA_PTR_HI) << 32) |
GEM_BANK1_READ_4(sc, GEM_TX_DATA_PTR_LO),
GEM_BANK1_READ_4(sc, GEM_TX_COMPLETION));
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
if (progress) {
if (sc->sc_txfree == GEM_NTXDESC - 1)
sc->sc_txwin = 0;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/*
* We freed some descriptors, so reset IFF_DRV_OACTIVE
* and restart.
*/
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
if (STAILQ_EMPTY(&sc->sc_txdirtyq))
sc->sc_wdog_timer = 0;
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
gem_start_locked(ifp);
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
}
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
CTR3(KTR_GEM, "%s: %s: watchdog %d",
device_get_name(sc->sc_dev), __func__, sc->sc_wdog_timer);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
- Wrap the handler and associated code for collecting completed RX descriptors that are still marked owned in #ifdef GEM_RINT_TIMEOUT instead of #if 0 for convenience. - Remove stale code and comment about relying on the preset XIF config. - In case of a watchdog timeout call the init function instead of just the start function so the chip is properly reset. Merge from hme(4): - Convert to use bus_dmamap_load_mbuf_sg() for loading RX buffers. - Protect from a duplicate mbuf free panic in case the DMA engine hangs. Reviewed by: yongari Tested on: powerpc(grehan), sparc64 MFC after: 1 week
2005-07-24 18:12:31 +00:00
#ifdef GEM_RINT_TIMEOUT
In some cases, RX descriptors that are signalled to have been completed by the hardware are still marked as owned. Handle this by installing a timeout handler to collect this descriptor to avoid having received packets remain unhandled until the next one arrives.
2002-03-23 19:43:15 +00:00
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_rint_timeout(void *arg)
In some cases, RX descriptors that are signalled to have been completed by the hardware are still marked as owned. Handle this by installing a timeout handler to collect this descriptor to avoid having received packets remain unhandled until the next one arrives.
2002-03-23 19:43:15 +00:00
{
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
struct gem_softc *sc = arg;
In some cases, RX descriptors that are signalled to have been completed by the hardware are still marked as owned. Handle this by installing a timeout handler to collect this descriptor to avoid having received packets remain unhandled until the next one arrives.
2002-03-23 19:43:15 +00:00
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
GEM_LOCK_ASSERT(sc, MA_OWNED);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
gem_rint(sc);
In some cases, RX descriptors that are signalled to have been completed by the hardware are still marked as owned. Handle this by installing a timeout handler to collect this descriptor to avoid having received packets remain unhandled until the next one arrives.
2002-03-23 19:43:15 +00:00
}
Fix warning.
2002-07-24 01:09:38 +00:00
#endif
In some cases, RX descriptors that are signalled to have been completed by the hardware are still marked as owned. Handle this by installing a timeout handler to collect this descriptor to avoid having received packets remain unhandled until the next one arrives.
2002-03-23 19:43:15 +00:00
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_rint(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
struct mbuf *m;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
uint64_t rxstat;
uint32_t rxcomp;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
GEM_LOCK_ASSERT(sc, MA_OWNED);
- Wrap the handler and associated code for collecting completed RX descriptors that are still marked owned in #ifdef GEM_RINT_TIMEOUT instead of #if 0 for convenience. - Remove stale code and comment about relying on the preset XIF config. - In case of a watchdog timeout call the init function instead of just the start function so the chip is properly reset. Merge from hme(4): - Convert to use bus_dmamap_load_mbuf_sg() for loading RX buffers. - Protect from a duplicate mbuf free panic in case the DMA engine hangs. Reviewed by: yongari Tested on: powerpc(grehan), sparc64 MFC after: 1 week
2005-07-24 18:12:31 +00:00
#ifdef GEM_RINT_TIMEOUT
In some cases, RX descriptors that are signalled to have been completed by the hardware are still marked as owned. Handle this by installing a timeout handler to collect this descriptor to avoid having received packets remain unhandled until the next one arrives.
2002-03-23 19:43:15 +00:00
callout_stop(&sc->sc_rx_ch);
- Wrap the handler and associated code for collecting completed RX descriptors that are still marked owned in #ifdef GEM_RINT_TIMEOUT instead of #if 0 for convenience. - Remove stale code and comment about relying on the preset XIF config. - In case of a watchdog timeout call the init function instead of just the start function so the chip is properly reset. Merge from hme(4): - Convert to use bus_dmamap_load_mbuf_sg() for loading RX buffers. - Protect from a duplicate mbuf free panic in case the DMA engine hangs. Reviewed by: yongari Tested on: powerpc(grehan), sparc64 MFC after: 1 week
2005-07-24 18:12:31 +00:00
#endif
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
CTR2(KTR_GEM, "%s: %s", device_get_name(sc->sc_dev), __func__);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
/*
* Read the completion register once. This limits
* how long the following loop can execute.
*/
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
rxcomp = GEM_BANK1_READ_4(sc, GEM_RX_COMPLETION);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
CTR3(KTR_GEM, "%s: sc->sc_rxptr %d, complete %d",
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
__func__, sc->sc_rxptr, rxcomp);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
GEM_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
for (; sc->sc_rxptr != rxcomp;) {
m = sc->sc_rxsoft[sc->sc_rxptr].rxs_mbuf;
rxstat = GEM_DMA_READ(sc,
sc->sc_rxdescs[sc->sc_rxptr].gd_flags);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
if (rxstat & GEM_RD_OWN) {
- Wrap the handler and associated code for collecting completed RX descriptors that are still marked owned in #ifdef GEM_RINT_TIMEOUT instead of #if 0 for convenience. - Remove stale code and comment about relying on the preset XIF config. - In case of a watchdog timeout call the init function instead of just the start function so the chip is properly reset. Merge from hme(4): - Convert to use bus_dmamap_load_mbuf_sg() for loading RX buffers. - Protect from a duplicate mbuf free panic in case the DMA engine hangs. Reviewed by: yongari Tested on: powerpc(grehan), sparc64 MFC after: 1 week
2005-07-24 18:12:31 +00:00
#ifdef GEM_RINT_TIMEOUT
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/*
In some cases, RX descriptors that are signalled to have been completed by the hardware are still marked as owned. Handle this by installing a timeout handler to collect this descriptor to avoid having received packets remain unhandled until the next one arrives.
2002-03-23 19:43:15 +00:00
* The descriptor is still marked as owned, although
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
* it is supposed to have completed. This has been
* observed on some machines. Just exiting here
In some cases, RX descriptors that are signalled to have been completed by the hardware are still marked as owned. Handle this by installing a timeout handler to collect this descriptor to avoid having received packets remain unhandled until the next one arrives.
2002-03-23 19:43:15 +00:00
* might leave the packet sitting around until another
* one arrives to trigger a new interrupt, which is
* generally undesirable, so set up a timeout.
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
*/
In some cases, RX descriptors that are signalled to have been completed by the hardware are still marked as owned. Handle this by installing a timeout handler to collect this descriptor to avoid having received packets remain unhandled until the next one arrives.
2002-03-23 19:43:15 +00:00
callout_reset(&sc->sc_rx_ch, GEM_RXOWN_TICKS,
gem_rint_timeout, sc);
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
#endif
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
m = NULL;
goto kickit;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
if (rxstat & GEM_RD_BAD_CRC) {
Mechanically convert to if_inc_counter().
2014-09-18 20:21:46 +00:00
if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
device_printf(sc->sc_dev, "receive error: CRC error\n");
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
GEM_INIT_RXDESC(sc, sc->sc_rxptr);
m = NULL;
goto kickit;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
#ifdef GEM_DEBUG
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
if ((ifp->if_flags & IFF_DEBUG) != 0) {
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
printf(" rxsoft %p descriptor %d: ",
&sc->sc_rxsoft[sc->sc_rxptr], sc->sc_rxptr);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
printf("gd_flags: 0x%016llx\t",
(long long)GEM_DMA_READ(sc,
sc->sc_rxdescs[sc->sc_rxptr].gd_flags));
printf("gd_addr: 0x%016llx\n",
(long long)GEM_DMA_READ(sc,
sc->sc_rxdescs[sc->sc_rxptr].gd_addr));
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
#endif
/*
* Allocate a new mbuf cluster. If that fails, we are
* out of memory, and must drop the packet and recycle
* the buffer that's already attached to this descriptor.
*/
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if (gem_add_rxbuf(sc, sc->sc_rxptr) != 0) {
Mechanically convert to if_inc_counter().
2014-09-18 20:21:46 +00:00
if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
GEM_INIT_RXDESC(sc, sc->sc_rxptr);
m = NULL;
}
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
kickit:
/*
* Update the RX kick register. This register has to point
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
* to the descriptor after the last valid one (before the
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
* current batch) and for optimum performance should be
* incremented in multiples of 4 (the DMA engine fetches/
* updates descriptors in batches of 4).
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
*/
sc->sc_rxptr = GEM_NEXTRX(sc->sc_rxptr);
if ((sc->sc_rxptr % 4) == 0) {
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_CDSYNC(sc,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_RX_KICK,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
(sc->sc_rxptr + GEM_NRXDESC - 4) &
GEM_NRXDESC_MASK);
}
if (m == NULL) {
if (rxstat & GEM_RD_OWN)
break;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
continue;
}
Mechanically convert to if_inc_counter().
2014-09-18 20:21:46 +00:00
if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
m->m_data += ETHER_ALIGN; /* first byte offset */
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
m->m_pkthdr.rcvif = ifp;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
m->m_pkthdr.len = m->m_len = GEM_RD_BUFLEN(rxstat);
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
gem_rxcksum(m, rxstat);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/* Pass it on. */
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_UNLOCK(sc);
network interface driver changes: o don't strip the Ethernet header from inbound packets; pass packets up the stack intact (required significant changes to some drivers) o reference common definitions in net/ethernet.h (e.g. ETHER_ALIGN) o track ether_ifattach/ether_ifdetach API changes o track bpf changes (use BPF_TAP and BPF_MTAP) o track vlan changes (ifnet capabilities, revised processing scheme, etc.) o use if_input to pass packets "up" o call ether_ioctl for default handling of ioctls Reviewed by: many Approved by: re
2002-11-14 23:54:55 +00:00
(*ifp->if_input)(ifp, m);
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_LOCK(sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
CTR3(KTR_GEM, "%s: done sc->sc_rxptr %d, complete %d", __func__,
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
sc->sc_rxptr, GEM_BANK1_READ_4(sc, GEM_RX_COMPLETION));
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
static int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_add_rxbuf(struct gem_softc *sc, int idx)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
struct gem_rxsoft *rxs = &sc->sc_rxsoft[idx];
struct mbuf *m;
- Wrap the handler and associated code for collecting completed RX descriptors that are still marked owned in #ifdef GEM_RINT_TIMEOUT instead of #if 0 for convenience. - Remove stale code and comment about relying on the preset XIF config. - In case of a watchdog timeout call the init function instead of just the start function so the chip is properly reset. Merge from hme(4): - Convert to use bus_dmamap_load_mbuf_sg() for loading RX buffers. - Protect from a duplicate mbuf free panic in case the DMA engine hangs. Reviewed by: yongari Tested on: powerpc(grehan), sparc64 MFC after: 1 week
2005-07-24 18:12:31 +00:00
bus_dma_segment_t segs[1];
int error, nsegs;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
GEM_LOCK_ASSERT(sc, MA_OWNED);
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
if (m == NULL)
return (ENOBUFS);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#ifdef GEM_DEBUG
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* Bzero the packet to check DMA. */
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
memset(m->m_ext.ext_buf, 0, m->m_ext.ext_size);
#endif
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Stop pretending that that we can sync part of a dmamap: replace the GEM_CDTXSYNC and GEM_CDRXSYNC macros with GEM_CDSYNC to sync the complete control map, and combine syncs wherever possible to avoid the overhead. - Sync all maps before unloading them. - Remove a few syncs which should be unnecessary.
2003-01-21 17:17:01 +00:00
if (rxs->rxs_mbuf != NULL) {
bus_dmamap_sync(sc->sc_rdmatag, rxs->rxs_dmamap,
BUS_DMASYNC_POSTREAD);
- Convert to use bus_dmamap_load_mbuf(). - remove DPRINTF(), there is a CTR*() for any of them, and KTR is far more useful to debug this driver. - some cleanups; remove some unused code and definitions.
2003-01-06 22:09:01 +00:00
bus_dmamap_unload(sc->sc_rdmatag, rxs->rxs_dmamap);
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Stop pretending that that we can sync part of a dmamap: replace the GEM_CDTXSYNC and GEM_CDRXSYNC macros with GEM_CDSYNC to sync the complete control map, and combine syncs wherever possible to avoid the overhead. - Sync all maps before unloading them. - Remove a few syncs which should be unnecessary.
2003-01-21 17:17:01 +00:00
}
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Wrap the handler and associated code for collecting completed RX descriptors that are still marked owned in #ifdef GEM_RINT_TIMEOUT instead of #if 0 for convenience. - Remove stale code and comment about relying on the preset XIF config. - In case of a watchdog timeout call the init function instead of just the start function so the chip is properly reset. Merge from hme(4): - Convert to use bus_dmamap_load_mbuf_sg() for loading RX buffers. - Protect from a duplicate mbuf free panic in case the DMA engine hangs. Reviewed by: yongari Tested on: powerpc(grehan), sparc64 MFC after: 1 week
2005-07-24 18:12:31 +00:00
error = bus_dmamap_load_mbuf_sg(sc->sc_rdmatag, rxs->rxs_dmamap,
m, segs, &nsegs, BUS_DMA_NOWAIT);
if (error != 0) {
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
device_printf(sc->sc_dev,
"cannot load RS DMA map %d, error = %d\n", idx, error);
- Wrap the handler and associated code for collecting completed RX descriptors that are still marked owned in #ifdef GEM_RINT_TIMEOUT instead of #if 0 for convenience. - Remove stale code and comment about relying on the preset XIF config. - In case of a watchdog timeout call the init function instead of just the start function so the chip is properly reset. Merge from hme(4): - Convert to use bus_dmamap_load_mbuf_sg() for loading RX buffers. - Protect from a duplicate mbuf free panic in case the DMA engine hangs. Reviewed by: yongari Tested on: powerpc(grehan), sparc64 MFC after: 1 week
2005-07-24 18:12:31 +00:00
m_freem(m);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (error);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* If nsegs is wrong then the stack is corrupt. */
- Fix some style bugs and remove another banal comment missed in rev. 1.46. - Move the KASSERT on gem_add_rxbuf() to the right spot and add an equivalent one to gem_disable_tx().
2008-03-09 17:55:19 +00:00
KASSERT(nsegs == 1,
("%s: too many DMA segments (%d)", __func__, nsegs));
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
rxs->rxs_mbuf = m;
- Wrap the handler and associated code for collecting completed RX descriptors that are still marked owned in #ifdef GEM_RINT_TIMEOUT instead of #if 0 for convenience. - Remove stale code and comment about relying on the preset XIF config. - In case of a watchdog timeout call the init function instead of just the start function so the chip is properly reset. Merge from hme(4): - Convert to use bus_dmamap_load_mbuf_sg() for loading RX buffers. - Protect from a duplicate mbuf free panic in case the DMA engine hangs. Reviewed by: yongari Tested on: powerpc(grehan), sparc64 MFC after: 1 week
2005-07-24 18:12:31 +00:00
rxs->rxs_paddr = segs[0].ds_addr;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
bus_dmamap_sync(sc->sc_rdmatag, rxs->rxs_dmamap,
BUS_DMASYNC_PREREAD);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
GEM_INIT_RXDESC(sc, idx);
return (0);
}
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_eint(struct gem_softc *sc, u_int status)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
Mechanically convert to if_inc_counter().
2014-09-18 20:21:46 +00:00
if_inc_counter(sc->sc_ifp, IFCOUNTER_IERRORS, 1);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((status & GEM_INTR_RX_TAG_ERR) != 0) {
gem_reset_rxdma(sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return;
}
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
device_printf(sc->sc_dev, "%s: status 0x%x", __func__, status);
if ((status & GEM_INTR_BERR) != 0) {
if ((sc->sc_flags & GEM_PCI) != 0)
printf(", PCI bus error 0x%x\n",
GEM_BANK1_READ_4(sc, GEM_PCI_ERROR_STATUS));
else
printf(", SBus error 0x%x\n",
GEM_BANK1_READ_4(sc, GEM_SBUS_STATUS));
}
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_intr(void *v)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
struct gem_softc *sc = v;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
uint32_t status, status2;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_LOCK(sc);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
status = GEM_BANK1_READ_4(sc, GEM_STATUS);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
CTR4(KTR_GEM, "%s: %s: cplt %x, status %x",
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
device_get_name(sc->sc_dev), __func__,
(status >> GEM_STATUS_TX_COMPLETION_SHFT), (u_int)status);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
/*
* PCS interrupts must be cleared, otherwise no traffic is passed!
*/
if ((status & GEM_INTR_PCS) != 0) {
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
status2 =
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_READ_4(sc, GEM_MII_INTERRUP_STATUS) |
GEM_BANK1_READ_4(sc, GEM_MII_INTERRUP_STATUS);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((status2 & GEM_MII_INTERRUP_LINK) != 0)
device_printf(sc->sc_dev,
"%s: PCS link status changed\n", __func__);
}
if ((status & GEM_MAC_CONTROL_STATUS) != 0) {
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
status2 = GEM_BANK1_READ_4(sc, GEM_MAC_CONTROL_STATUS);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((status2 & GEM_MAC_PAUSED) != 0)
device_printf(sc->sc_dev,
"%s: PAUSE received (PAUSE time %d slots)\n",
__func__, GEM_MAC_PAUSE_TIME(status2));
if ((status2 & GEM_MAC_PAUSE) != 0)
device_printf(sc->sc_dev,
"%s: transited to PAUSE state\n", __func__);
if ((status2 & GEM_MAC_RESUME) != 0)
device_printf(sc->sc_dev,
"%s: transited to non-PAUSE state\n", __func__);
}
if ((status & GEM_INTR_MIF) != 0)
device_printf(sc->sc_dev, "%s: MIF interrupt\n", __func__);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
if (__predict_false(status &
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
(GEM_INTR_RX_TAG_ERR | GEM_INTR_PERR | GEM_INTR_BERR)) != 0)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
gem_eint(sc, status);
if ((status & (GEM_INTR_RX_DONE | GEM_INTR_RX_NOBUF)) != 0)
gem_rint(sc);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((status & (GEM_INTR_TX_EMPTY | GEM_INTR_TX_INTME)) != 0)
gem_tint(sc);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
if (__predict_false((status & GEM_INTR_TX_MAC) != 0)) {
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
status2 = GEM_BANK1_READ_4(sc, GEM_MAC_TX_STATUS);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
if ((status2 &
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
~(GEM_MAC_TX_XMIT_DONE | GEM_MAC_TX_DEFER_EXP |
GEM_MAC_TX_PEAK_EXP)) != 0)
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
device_printf(sc->sc_dev,
"MAC TX fault, status %x\n", status2);
if ((status2 &
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
(GEM_MAC_TX_UNDERRUN | GEM_MAC_TX_PKT_TOO_LONG)) != 0) {
Mechanically convert to if_inc_counter().
2014-09-18 20:21:46 +00:00
if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
Remove unnecessary controller reinitialization by checking IFF_DRV_RUNNING flag. Previously running dhclient or adding alias addresses reinitialized controller and it resulted in unnecessary link flips. Reviewed by: marius
2011-05-20 17:01:22 +00:00
sc->sc_ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
gem_init_locked(sc);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
}
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
if (__predict_false((status & GEM_INTR_RX_MAC) != 0)) {
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
status2 = GEM_BANK1_READ_4(sc, GEM_MAC_RX_STATUS);
- Allow for VLAN-sized frames and set IFCAP_VLAN_MTU. - On resume all registers have to be initialized again like after power-on so reset sc_inited in gem_suspend() in order get all of the registers set next time gem_init_regs() is called. - On at least some ERI and GEM revisions GEM_MAC_RX_OVERFLOW happen often due to a silicon bug and re-initializing is all we can do about these errors so make handling them non-verbose. - Remove a superfluous memset(3) call in gem_meminit(), all elements are initialized to 0 anyway. MFC after: 1 week
2005-08-28 15:07:30 +00:00
/*
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
* At least with GEM_SUN_GEM and some GEM_SUN_ERI
* revisions GEM_MAC_RX_OVERFLOW happen often due to a
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
* silicon bug so handle them silently. Moreover, it's
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
* likely that the receiver has hung so we reset it.
- Allow for VLAN-sized frames and set IFCAP_VLAN_MTU. - On resume all registers have to be initialized again like after power-on so reset sc_inited in gem_suspend() in order get all of the registers set next time gem_init_regs() is called. - On at least some ERI and GEM revisions GEM_MAC_RX_OVERFLOW happen often due to a silicon bug and re-initializing is all we can do about these errors so make handling them non-verbose. - Remove a superfluous memset(3) call in gem_meminit(), all elements are initialized to 0 anyway. MFC after: 1 week
2005-08-28 15:07:30 +00:00
*/
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
if ((status2 & GEM_MAC_RX_OVERFLOW) != 0) {
Mechanically convert to if_inc_counter().
2014-09-18 20:21:46 +00:00
if_inc_counter(sc->sc_ifp, IFCOUNTER_IERRORS, 1);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
gem_reset_rxdma(sc);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
} else if ((status2 &
~(GEM_MAC_RX_DONE | GEM_MAC_RX_FRAME_CNT)) != 0)
device_printf(sc->sc_dev,
"MAC RX fault, status %x\n", status2);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_UNLOCK(sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
- Use the gem_tick() callout instead of if_slowtimo() for driving gem_watchdog() in order to avoid races accessing if_timer. While at it relax the watchdog a bit by reloading it in gem_tint() if there are still packets enqueued. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Fix inconsistencies in prototypes.
2006-12-06 02:04:25 +00:00
static int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_watchdog(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Use the gem_tick() callout instead of if_slowtimo() for driving gem_watchdog() in order to avoid races accessing if_timer. While at it relax the watchdog a bit by reloading it in gem_tint() if there are still packets enqueued. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Fix inconsistencies in prototypes.
2006-12-06 02:04:25 +00:00
GEM_LOCK_ASSERT(sc, MA_OWNED);
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#ifdef GEM_DEBUG
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
CTR4(KTR_GEM,
"%s: GEM_RX_CONFIG %x GEM_MAC_RX_STATUS %x GEM_MAC_RX_CONFIG %x",
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
__func__, GEM_BANK1_READ_4(sc, GEM_RX_CONFIG),
GEM_BANK1_READ_4(sc, GEM_MAC_RX_STATUS),
GEM_BANK1_READ_4(sc, GEM_MAC_RX_CONFIG));
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
CTR4(KTR_GEM,
"%s: GEM_TX_CONFIG %x GEM_MAC_TX_STATUS %x GEM_MAC_TX_CONFIG %x",
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
__func__, GEM_BANK1_READ_4(sc, GEM_TX_CONFIG),
GEM_BANK1_READ_4(sc, GEM_MAC_TX_STATUS),
GEM_BANK1_READ_4(sc, GEM_MAC_TX_CONFIG));
Miscellaneous fixes: - Fix compilation without GEM_DEBUG. - Do not #define GEM_DEBUG by default; it adds overhead (due to bzero()ing RX space) and is not needed any more, since the driver is quite stable now. - Fix watchdog timeouts when failing to load TX packets. - Do not forcibly limit the number of descriptors used for a packet to GEM_NTXSEGS, by passing this number to bus_dma_tag_create(). There is no requirement for a limit any lower than the total number of available descriptors, and the present limit caused network problems due to mbuf chains requiring more descriptors. GEM_NTXSEGS is still used to estimate the interrupt window size, for which we just need an estimate. Approved by: re (rwatson)
2003-05-15 16:57:55 +00:00
#endif
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- Use the gem_tick() callout instead of if_slowtimo() for driving gem_watchdog() in order to avoid races accessing if_timer. While at it relax the watchdog a bit by reloading it in gem_tint() if there are still packets enqueued. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Fix inconsistencies in prototypes.
2006-12-06 02:04:25 +00:00
if (sc->sc_wdog_timer == 0 || --sc->sc_wdog_timer != 0)
return (0);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((sc->sc_flags & GEM_LINK) != 0)
device_printf(sc->sc_dev, "device timeout\n");
else if (bootverbose)
device_printf(sc->sc_dev, "device timeout (no link)\n");
Mechanically convert to if_inc_counter().
2014-09-18 20:21:46 +00:00
if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/* Try to get more packets going. */
Remove unnecessary controller reinitialization by checking IFF_DRV_RUNNING flag. Previously running dhclient or adding alias addresses reinitialized controller and it resulted in unnecessary link flips. Reviewed by: marius
2011-05-20 17:01:22 +00:00
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
gem_init_locked(sc);
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
gem_start_locked(ifp);
- Use the gem_tick() callout instead of if_slowtimo() for driving gem_watchdog() in order to avoid races accessing if_timer. While at it relax the watchdog a bit by reloading it in gem_tint() if there are still packets enqueued. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Fix inconsistencies in prototypes.
2006-12-06 02:04:25 +00:00
return (EJUSTRETURN);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
static void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_mifinit(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
- Fix some style bugs and remove another banal comment missed in rev. 1.46. - Move the KASSERT on gem_add_rxbuf() to the right spot and add an equivalent one to gem_disable_tx().
2008-03-09 17:55:19 +00:00
/* Configure the MIF in frame mode. */
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MIF_CONFIG,
GEM_BANK1_READ_4(sc, GEM_MIF_CONFIG) & ~GEM_MIF_CONFIG_BB_ENA);
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
GEM_BANK1_BARRIER(sc, GEM_MIF_CONFIG, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
/*
* MII interface
*
- Restore the behavior of enabling the MII buffer for an internal PHY only and not also in the case of an external PHY currently doing full duplex, which accidentally got broken in r172334. It's still not clear to me why we need to enable the buffer for an internal PHY though. - Count excess and late collisions as output errors. [1] - Count receive errors as input errors. [1] Obtained from: NetBSD [1] MFC after: 3 days
2008-08-23 15:03:26 +00:00
* The MII interface supports at least three different operating modes:
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
*
* Bitbang mode is implemented using data, clock and output enable registers.
*
* Frame mode is implemented by loading a complete frame into the frame
* register and polling the valid bit for completion.
*
* Polling mode uses the frame register but completion is indicated by
* an interrupt.
*
*/
int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_mii_readreg(device_t dev, int phy, int reg)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
struct gem_softc *sc;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
int n;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
uint32_t v;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#ifdef GEM_DEBUG_PHY
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
printf("%s: phy %d reg %d\n", __func__, phy, reg);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#endif
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
sc = device_get_softc(dev);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((sc->sc_flags & GEM_SERDES) != 0) {
switch (reg) {
case MII_BMCR:
reg = GEM_MII_CONTROL;
break;
case MII_BMSR:
reg = GEM_MII_STATUS;
break;
case MII_PHYIDR1:
case MII_PHYIDR2:
return (0);
case MII_ANAR:
reg = GEM_MII_ANAR;
break;
case MII_ANLPAR:
reg = GEM_MII_ANLPAR;
break;
case MII_EXTSR:
return (EXTSR_1000XFDX | EXTSR_1000XHDX);
default:
device_printf(sc->sc_dev,
"%s: unhandled register %d\n", __func__, reg);
return (0);
}
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
return (GEM_BANK1_READ_4(sc, reg));
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
}
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* Construct the frame command. */
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
v = GEM_MIF_FRAME_READ |
(phy << GEM_MIF_PHY_SHIFT) |
(reg << GEM_MIF_REG_SHIFT);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MIF_FRAME, v);
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_BANK1_BARRIER(sc, GEM_MIF_FRAME, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
for (n = 0; n < 100; n++) {
DELAY(1);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
v = GEM_BANK1_READ_4(sc, GEM_MIF_FRAME);
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
if (v & GEM_MIF_FRAME_TA0)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return (v & GEM_MIF_FRAME_DATA);
}
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
device_printf(sc->sc_dev, "%s: timed out\n", __func__);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return (0);
}
int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_mii_writereg(device_t dev, int phy, int reg, int val)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
struct gem_softc *sc;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
int n;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
uint32_t v;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#ifdef GEM_DEBUG_PHY
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
printf("%s: phy %d reg %d val %x\n", phy, reg, val, __func__);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#endif
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
sc = device_get_softc(dev);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((sc->sc_flags & GEM_SERDES) != 0) {
switch (reg) {
case MII_BMSR:
reg = GEM_MII_STATUS;
break;
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
case MII_BMCR:
reg = GEM_MII_CONTROL;
if ((val & GEM_MII_CONTROL_RESET) == 0)
break;
GEM_BANK1_WRITE_4(sc, GEM_MII_CONTROL, val);
GEM_BANK1_BARRIER(sc, GEM_MII_CONTROL, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
if (!GEM_BANK1_BITWAIT(sc, GEM_MII_CONTROL,
GEM_MII_CONTROL_RESET, 0))
device_printf(sc->sc_dev,
"cannot reset PCS\n");
/* FALLTHROUGH */
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
case MII_ANAR:
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MII_CONFIG, 0);
GEM_BANK1_BARRIER(sc, GEM_MII_CONFIG, 4,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
BUS_SPACE_BARRIER_WRITE);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MII_ANAR, val);
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
GEM_BANK1_BARRIER(sc, GEM_MII_ANAR, 4,
BUS_SPACE_BARRIER_WRITE);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MII_SLINK_CONTROL,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
GEM_MII_SLINK_LOOPBACK | GEM_MII_SLINK_EN_SYNC_D);
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
GEM_BANK1_BARRIER(sc, GEM_MII_SLINK_CONTROL, 4,
BUS_SPACE_BARRIER_WRITE);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MII_CONFIG,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
GEM_MII_CONFIG_ENABLE);
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
GEM_BANK1_BARRIER(sc, GEM_MII_CONFIG, 4,
BUS_SPACE_BARRIER_WRITE);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (0);
case MII_ANLPAR:
reg = GEM_MII_ANLPAR;
break;
default:
device_printf(sc->sc_dev,
"%s: unhandled register %d\n", __func__, reg);
return (0);
}
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, reg, val);
Merge r207585 from cas(4): - Don't probe for PHYs if we already know to use a SERDES. Unlike as with cas(4) this only serves to speed up the the device attach though and can only be determined via the OFW device tree but not from the VPD. - Don't touch the MIF when using a SERDES. - Add some missing bus space barriers, mainly in the PCS code path.
2010-09-16 09:29:48 +00:00
GEM_BANK1_BARRIER(sc, reg, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
return (0);
}
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* Construct the frame command. */
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
v = GEM_MIF_FRAME_WRITE |
(phy << GEM_MIF_PHY_SHIFT) |
(reg << GEM_MIF_REG_SHIFT) |
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
(val & GEM_MIF_FRAME_DATA);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MIF_FRAME, v);
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_BANK1_BARRIER(sc, GEM_MIF_FRAME, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
for (n = 0; n < 100; n++) {
DELAY(1);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
v = GEM_BANK1_READ_4(sc, GEM_MIF_FRAME);
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
if (v & GEM_MIF_FRAME_TA0)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return (1);
}
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
device_printf(sc->sc_dev, "%s: timed out\n", __func__);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return (0);
}
void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_mii_statchg(device_t dev)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
struct gem_softc *sc;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
int gigabit;
uint32_t rxcfg, txcfg, v;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
sc = device_get_softc(dev);
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
GEM_LOCK_ASSERT(sc, MA_OWNED);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#ifdef GEM_DEBUG
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
if ((sc->sc_ifp->if_flags & IFF_DEBUG) != 0)
Convert the PHY drivers to honor the mii_flags passed down and convert the NIC drivers as well as the PHY drivers to take advantage of the mii_attach() introduced in r213878 to get rid of certain hacks. For the most part these were: - Artificially limiting miibus_{read,write}reg methods to certain PHY addresses; we now let mii_attach() only probe the PHY at the desired address(es) instead. - PHY drivers setting MIIF_* flags based on the NIC driver they hang off from, partly even based on grabbing and using the softc of the parent; we now pass these flags down from the NIC to the PHY drivers via mii_attach(). This got us rid of all such hacks except those of brgphy() in combination with bce(4) and bge(4), which is way beyond what can be expressed with simple flags. While at it, I took the opportunity to change the NIC drivers to pass up the error returned by mii_attach() (previously by mii_phy_probe()) and unify the error message used in this case where and as appropriate as mii_attach() actually can fail for a number of reasons, not just because of no PHY(s) being present at the expected address(es). Reviewed by: jhb, yongari
2010-10-15 14:52:11 +00:00
device_printf(sc->sc_dev, "%s: status change\n", __func__);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
#endif
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((sc->sc_mii->mii_media_status & IFM_ACTIVE) != 0 &&
IFM_SUBTYPE(sc->sc_mii->mii_media_active) != IFM_NONE)
sc->sc_flags |= GEM_LINK;
else
sc->sc_flags &= ~GEM_LINK;
switch (IFM_SUBTYPE(sc->sc_mii->mii_media_active)) {
case IFM_1000_SX:
case IFM_1000_LX:
case IFM_1000_CX:
case IFM_1000_T:
gigabit = 1;
break;
default:
gigabit = 0;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
/*
* The configuration done here corresponds to the steps F) and
* G) and as far as enabling of RX and TX MAC goes also step H)
* of the initialization sequence outlined in section 3.2.1 of
* the GEM Gigabit Ethernet ASIC Specification.
*/
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
rxcfg = sc->sc_mac_rxcfg;
rxcfg &= ~GEM_MAC_RX_CARR_EXTEND;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
txcfg = GEM_MAC_TX_ENA_IPG0 | GEM_MAC_TX_NGU | GEM_MAC_TX_NGU_LIMIT;
if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
txcfg |= GEM_MAC_TX_IGN_CARRIER | GEM_MAC_TX_IGN_COLLIS;
else if (gigabit != 0) {
rxcfg |= GEM_MAC_RX_CARR_EXTEND;
txcfg |= GEM_MAC_TX_CARR_EXTEND;
}
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
(void)gem_disable_tx(sc);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_TX_CONFIG, txcfg);
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
(void)gem_disable_rx(sc);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_CONFIG, rxcfg);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
v = GEM_BANK1_READ_4(sc, GEM_MAC_CONTROL_CONFIG) &
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
~(GEM_MAC_CC_RX_PAUSE | GEM_MAC_CC_TX_PAUSE);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
IFM_ETH_RXPAUSE) != 0)
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
v |= GEM_MAC_CC_RX_PAUSE;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
IFM_ETH_TXPAUSE) != 0)
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
v |= GEM_MAC_CC_TX_PAUSE;
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_CONTROL_CONFIG, v);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0 &&
gigabit != 0)
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_SLOT_TIME,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
GEM_MAC_SLOT_TIME_CARR_EXTEND);
else
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_SLOT_TIME,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
GEM_MAC_SLOT_TIME_NORMAL);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
/* XIF Configuration */
v = GEM_MAC_XIF_LINK_LED;
v |= GEM_MAC_XIF_TX_MII_ENA;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((sc->sc_flags & GEM_SERDES) == 0) {
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
if ((GEM_BANK1_READ_4(sc, GEM_MIF_CONFIG) &
- Restore the behavior of enabling the MII buffer for an internal PHY only and not also in the case of an external PHY currently doing full duplex, which accidentally got broken in r172334. It's still not clear to me why we need to enable the buffer for an internal PHY though. - Count excess and late collisions as output errors. [1] - Count receive errors as input errors. [1] Obtained from: NetBSD [1] MFC after: 3 days
2008-08-23 15:03:26 +00:00
GEM_MIF_CONFIG_PHY_SEL) != 0) {
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
/* External MII needs echo disable if half duplex. */
- Initialize the ifnet structure, especially if_dname, before probing the PHYs as some PHY drivers use it (but probably shouldn't). How gem(4) has worked with brgphy(4) on powerpc without this so far is unclear to me. - Introduce a dying flag which is set during detach and checked in gem_ioctl() in order to prevent active BPF listeners to clear promiscuous mode which may lead to the tick callout being restarted which will trigger a panic once it's actually gone. - In gem_stop() reset rather than just disable the transmitter and receiver in order to ensure we're not unloading DMA maps still in use by the hardware. [1] - The blanking time is specified in PCI clocks so we should use twice the value when operating at 66MHz. - Spell some 2 as ETHER_ALIGN and a 19 as GEM_STATUS_TX_COMPLETION_SHFT to make the actual intentions clear. - As we don't unload the peak attempts counter ignore its overflow interrupts. - Remove a stale setting of a variable to GEM_TD_INTERRUPT_ME which isn't used afterwards. - For optimum performance increment the TX kick register in multiples of 4 if possible as suggested by the documentation. - Partially revert r164931; drivers should only clear the watchdog timer if all outstanding TX descriptors are done. - Fix some debugging strings. - Add a missing BUS_DMASYNC_POSTWRITE in gem_rint(). - As the error paths in the interrupt handler are generally unlikely predict them as false. - Add support for the SBus version of the GEM controller. [2] - Add some lock assertions. - Improve some comments. - Fix some more or less cosmetic issues in the code of the PCI front-end. - Change some softc members to be unsigned where more appropriate and remove unused ones. Approved by: re (kib) Obtained from: NetBSD (partially) [2], OpenBSD [1] MFC after: 2 weeks
2009-06-23 20:36:59 +00:00
if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
- Restore the behavior of enabling the MII buffer for an internal PHY only and not also in the case of an external PHY currently doing full duplex, which accidentally got broken in r172334. It's still not clear to me why we need to enable the buffer for an internal PHY though. - Count excess and late collisions as output errors. [1] - Count receive errors as input errors. [1] Obtained from: NetBSD [1] MFC after: 3 days
2008-08-23 15:03:26 +00:00
IFM_FDX) == 0)
v |= GEM_MAC_XIF_ECHO_DISABL;
} else
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
/*
* Internal MII needs buffer enable.
* XXX buffer enable makes only sense for an
* external PHY.
*/
v |= GEM_MAC_XIF_MII_BUF_ENA;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if (gigabit != 0)
v |= GEM_MAC_XIF_GMII_MODE;
if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
v |= GEM_MAC_XIF_FDPLX_LED;
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_XIF_CONFIG, v);
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
sc->sc_mac_rxcfg = rxcfg;
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
(sc->sc_flags & GEM_LINK) != 0) {
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_TX_CONFIG,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
txcfg | GEM_MAC_TX_ENABLE);
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_CONFIG,
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
rxcfg | GEM_MAC_RX_ENABLE);
}
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_mediachange(struct ifnet *ifp)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
struct gem_softc *sc = ifp->if_softc;
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
int error;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* XXX add support for serial media. */
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
GEM_LOCK(sc);
error = mii_mediachg(sc->sc_mii);
GEM_UNLOCK(sc);
return (error);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
void
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
struct gem_softc *sc = ifp->if_softc;
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_LOCK(sc);
if ((ifp->if_flags & IFF_UP) == 0) {
GEM_UNLOCK(sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
return;
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
}
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
mii_pollstat(sc->sc_mii);
ifmr->ifm_active = sc->sc_mii->mii_media_active;
ifmr->ifm_status = sc->sc_mii->mii_media_status;
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_UNLOCK(sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
static int
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
gem_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
struct gem_softc *sc = ifp->if_softc;
struct ifreq *ifr = (struct ifreq *)data;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
int error;
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
error = 0;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
switch (cmd) {
case SIOCSIFFLAGS:
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
GEM_LOCK(sc);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
if ((ifp->if_flags & IFF_UP) != 0) {
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
((ifp->if_flags ^ sc->sc_ifflags) &
(IFF_ALLMULTI | IFF_PROMISC)) != 0)
Correct r223648; as gem_init_locked() was calling gem_setladrf(), which sets GEM_MAC_RX_CONFIG based on sc_mac_rxcfg which in turn is initialized to zero, before reading the supposedly default configuration we were effectively not basing sc_mac_rxcfg and thus GEM_MAC_RX_CONFIG on the default configuration. Solve this by calling gem_setladrf() after reading in the default configuration of GEM_MAC_RX_CONFIG. This also avoids the need to distinguish whether gem_setladrf() should enable the RX MAC again and should be slightly more correct as we're now doing all of the RX MAC configuration in the intended step.
2011-07-12 13:06:44 +00:00
gem_setladrf(sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
else
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
gem_init_locked(sc);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
} else if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
gem_stop(ifp, 0);
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
if ((ifp->if_flags & IFF_LINK0) != 0)
sc->sc_csum_features |= CSUM_UDP;
else
sc->sc_csum_features &= ~CSUM_UDP;
if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
ifp->if_hwassist = sc->sc_csum_features;
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
sc->sc_ifflags = ifp->if_flags;
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
GEM_UNLOCK(sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
GEM_LOCK(sc);
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
Correct r223648; as gem_init_locked() was calling gem_setladrf(), which sets GEM_MAC_RX_CONFIG based on sc_mac_rxcfg which in turn is initialized to zero, before reading the supposedly default configuration we were effectively not basing sc_mac_rxcfg and thus GEM_MAC_RX_CONFIG on the default configuration. Solve this by calling gem_setladrf() after reading in the default configuration of GEM_MAC_RX_CONFIG. This also avoids the need to distinguish whether gem_setladrf() should enable the RX MAC again and should be slightly more correct as we're now doing all of the RX MAC configuration in the intended step.
2011-07-12 13:06:44 +00:00
gem_setladrf(sc);
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
GEM_UNLOCK(sc);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
break;
case SIOCGIFMEDIA:
case SIOCSIFMEDIA:
error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii->mii_media, cmd);
break;
o Implemented Rx/Tx checksum offload. The simple checksum logic in GEMs is unable to discriminate UDP from TCP packets such that it can generate 0x0000 checksum value for the UDP datagram. So the UDP checksum offload was disabled by default. You can enable it by setting link0 flag with ifconfig(8). o bus_dma(9) clean up. It now correctly set number of required DMA segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag in static allocations done via bus_dmamem_alloc(9). o Implemented ALTQ(9) support. o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove several book keeping chores orginated from call-back mechanism. Therefore gem_txdma_callback() was removed and its functionality was reimplemented in gem_load_txmbuf(). o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf chains are set. I think it was a long standing bug and it caused fluctuating interrupts/CPU usage patterns while netperf test is in progress. Previously it seems that we race with the device. Because I don't have a documentation for GEM I'm not sure this is correct but almost all other documentations I have stated this implications on setting SOP mark in descriptor ring(e.g. hme(4)). o Borrowed gem_defrag() from ath(4) which is supposed to be much faster than m_defrag(9) since it's not need to defrag all mbuf chains. o gem_load_txmbuf() was changed to allow passed mbuf chains to free. Caller of gem_load_txmbuf() correctly handles freed mbuf chains. o In gem_start_locked(), added checks for availability of Tx descriptors before trying to load DMA maps which could save CPU cycles when number of available descriptors are low. Also, simplyfy IFF_DRV_OACTIVE detection logic. o Removed hard-coded function names in CTR macros and replaced it with __func__. o Moved statistics counter register access to gem_tick() to reduce number of PCI bus accesses. There is no reason to update statistics counters in interrupt handler. o Removed unnecessary call of gem_start_locked() in gem_ioctl(). Reviewed by: grehan (initial version), marius (with improvements and suggestions) Tested by: grehan (ppc), marius(sparc64)
2007-06-04 06:01:04 +00:00
case SIOCSIFCAP:
GEM_LOCK(sc);
ifp->if_capenable = ifr->ifr_reqcap;
if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
ifp->if_hwassist = sc->sc_csum_features;
else
ifp->if_hwassist = 0;
GEM_UNLOCK(sc);
break;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
default:
- In gem_ioctl() move the call to ether_ioctl() to the default case of the switch statement in order to make this driver more like other Ethernet NIC drivers. - In gem_attach() call gem_stop() in addition to gem_reset() to make sure the chip actually is stopped and not just reset. - In gem_stop() also stop the gem_rint_timeout() callout in case the driver is compiled with GEM_RINT_TIMEOUT defined. Merge some locking improvements from hme(4): - Use callout_init_mtx() to close races between gem_stop() and gem_tick() as weel as gem_stop() and gem_rint() in case the driver is compiled with GEM_RINT_TIMEOUT defined. - Use the driver lock instead of Giant in a bus dma callback. - Lock the driver lock around mii operations. - Cleanup locking in gem_ioctl(). - Remove redundant assertions that the driver lock is not held in gem_attach() and gem_detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Add callout_drain()'s to gem_detach() after calling gem_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex.
2005-09-18 13:23:19 +00:00
error = ether_ioctl(ifp, cmd, data);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
break;
}
return (error);
}
Convert to if_foreach_llmaddr() KPI.
2019-10-21 18:06:57 +00:00
static u_int
gem_hash_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
{
uint32_t crc, *hash = arg;
crc = ether_crc32_le(LLADDR(sdl), ETHER_ADDR_LEN);
/* We just want the 8 most significant bits. */
crc >>= 24;
/* Set the corresponding bit in the filter. */
hash[crc >> 4] |= 1 << (15 - (crc & 15));
return (1);
}
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
static void
Correct r223648; as gem_init_locked() was calling gem_setladrf(), which sets GEM_MAC_RX_CONFIG based on sc_mac_rxcfg which in turn is initialized to zero, before reading the supposedly default configuration we were effectively not basing sc_mac_rxcfg and thus GEM_MAC_RX_CONFIG on the default configuration. Solve this by calling gem_setladrf() after reading in the default configuration of GEM_MAC_RX_CONFIG. This also avoids the need to distinguish whether gem_setladrf() should enable the RX MAC again and should be slightly more correct as we're now doing all of the RX MAC configuration in the intended step.
2011-07-12 13:06:44 +00:00
gem_setladrf(struct gem_softc *sc)
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
int i;
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
uint32_t hash[16];
Convert to if_foreach_llmaddr() KPI.
2019-10-21 18:06:57 +00:00
uint32_t v;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Remove NOP spl*() calls and add locking (making gem(4) MPSAFE). Based on: hme(4) Reviewed by: yongari Tested on: powerpc(grehan), sparc64
2005-07-24 18:45:15 +00:00
GEM_LOCK_ASSERT(sc, MA_OWNED);
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
/*
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
* Turn off the RX MAC and the hash filter as required by the Sun GEM
* programming restrictions.
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
*/
Fix typo in r223648 which was accidentally committed
2011-06-28 16:44:02 +00:00
v = sc->sc_mac_rxcfg & ~GEM_MAC_RX_HASH_FILTER;
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_CONFIG, v);
o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Given that gem_load_txmbuf() is allowed to fail resulting in a packet drop also for quite some time now implement the functionality of gem_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in gem_rint() and gem_start_locked(). o Correct some bus_barrier(9) calls to do a read/write barrier as we do a read after a write. Add some missing ones in gem_mii_readreg() and gem_mii_writereg(). o According to the Apple GMAC driver, the GEM ASIC specification and the OpenSolaris eri(7D) the TX FIFO threshold has to be set to 0x4ff for the Gigabit variants and 0x100 for the ERI in order do avoid TX underruns. o In gem_init_locked(): - be conservative and enable the RX and TX MACs, - don't clear GEM_LINK otherwise we don't ever mark the link as up again if gem_init_locked() is called from gem_watchdog(), - remove superfluous setting of sc_ifflags. o Don't bother to check whether the interface is running or whether its queue is empty before calling gem_start_locked() in gem_tint(), the former will check these anyway. o Call gem_start_locked() in gem_watchdog() in order to try to get some more packets going. o In gem_mii_writereg() after reseting the PCS restore its configuration. GMAC testing: grehan, marcel MFC after: 2 weeks
2008-06-22 13:54:51 +00:00
GEM_BANK1_BARRIER(sc, GEM_MAC_RX_CONFIG, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
if (!GEM_BANK1_BITWAIT(sc, GEM_MAC_RX_CONFIG, GEM_MAC_RX_HASH_FILTER |
GEM_MAC_RX_ENABLE, 0))
device_printf(sc->sc_dev,
"cannot disable RX MAC or hash filter\n");
o Revert the part of if_gem.c rev. 1.35 which added a call to gem_stop() to gem_attach() as the former access softc members not yet initialized at that time and gem_reset() actually is enough to stop the chip. [1] o Revise the use of gem_bitwait(); add bus_barrier() calls before calling gem_bitwait() to ensure the respective bit has been written before we starting polling on it and poll for the right bits to change, f.e. even though we only reset RX we have to actually wait for both GEM_RESET_RX and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in places we've been missing them according to the GEM documentation. Along with this some excessive DELAYs, which probably only were added because of bugs in gem_bitwait() and its use in the first place, as well as as have of an gem_bitwait() reimplementation in gem_reset_tx() were removed. o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors more gracefully as unlike gem_init_locked() it resets the RX DMA engine only, causing no link loss and the FIFOs not to be cleared. Also use it deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled. This was based on information obtained from the Linux GEM and OpenSolaris ERI drivers. o Turn on workarounds for silicon bugs in the Apple GMAC variants. This was based on information obtained from the Darwin GMAC and Linux GEM drivers. o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts. This greatly improves especially RX performance. o Optimize the RX path, this consists of: - kicking the receiver as soon as we've a spare descriptor in gem_rint() again instead of just once after all the ready ones have been handled; - kicking the receiver the right way, i.e. as outlined in the GEM documentation in batches of 4 and by pointing it to the descriptor after the last valid one; - calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may take quite a while; - doubling the size of the RX ring to 256 descriptors. Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved from ~100Mbit/s to ~850Mbit/s. o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg() fails we'll free the newly allocated mbuf, unable to recycle the previous one but a NULL pointer dereference instead. o In gem_init_locked() honor the return value of gem_meminit(). o Simplify gem_ringsize() and dont' return garbage in the default case. Based on OpenBSD. o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is defined as we don't need/use these interrupts for operation. o In gem_start_locked() sync the DMA maps of the descriptor rings before every kick of the transmitter and not just once after enqueuing all packets as the NIC might instantly start transmitting after we kicked it the first time. o Keep state of the link state and use it to enable or disable the MAC in gem_mii_statchg() accordingly as well as to return early from gem_start_locked() in case the link is down. [3] o Initialize the maximum frame size to a sane value. o In gem_mii_statchg() enable carrier extension if appropriate. o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in gem_eint(). [3] o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous group mode on and don't clear the flag if we've disabled promiscuous group mode (these were mostly NOPs though). [2] o Let gem_eint() also report GEM_INTR_PERR errors. o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as device probe methods are not supposed to touch the softc. o Collapse sc_inited and sc_pci into bits for sc_flags. o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to legal values. o Correctly set up for 802.3x flow control, though #ifdef out the code that actually enables it as this needs more testing and mainly a proper framework to support it. o Correct and add some conversions from hard-coded functions names to __func__ which were borked or forgotten in if_gem.c rev. 1.42. o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o In gem_pci_attach() in case attaching fails release the resources in the opposite order they were allocated. o Make gem_reset() static to if_gem.c as it's not needed outside that module. o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was never set and the associated code was in the wrong place. o Remove sc_mif_config; it was only used to cache the contents of the respective register within gem_attach(). o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend hook as it will never be used on FreeBSD. o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for Apple K2 GMAC. Based on OpenBSD. o Add support for Sun GBE/P cards, or in other words actually add support for cards based on GEM to gem(4). This mainly consists of adding support for the TBI of these chips. Along with this the PHY selection code was rewritten to hardcode the PHY number for certain configurations as for example the PHY of the on-board ERI of Blade 1000 shows up twice causing no link as the second incarnation is isolated. These changes were ported from OpenBSD with some additional improvements and modulo some bugs. o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on systems without Open Firmware. This is an improved version of my variant of the respective code in if_hme_pci.c o Now that gem(4) is MI enable it for all archs. Pointed out by: yongari [1] Suggested by: rwatson [2], yongari [3] Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari), sparc64 (ERI and GEM) Reviewed by: yongari Approved by: re (kensmith)
2007-09-26 21:14:18 +00:00
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
v &= ~(GEM_MAC_RX_PROMISCUOUS | GEM_MAC_RX_PROMISC_GRP);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
if ((ifp->if_flags & IFF_PROMISC) != 0) {
v |= GEM_MAC_RX_PROMISCUOUS;
goto chipit;
}
if ((ifp->if_flags & IFF_ALLMULTI) != 0) {
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
v |= GEM_MAC_RX_PROMISC_GRP;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
goto chipit;
}
/*
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
* Set up multicast address filter by passing all multicast
* addresses through a crc generator, and then using the high
* order 8 bits as an index into the 256 bit logical address
* filter. The high order 4 bits selects the word, while the
* other 4 bits select the bit within the word (where bit 0
* is the MSB).
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
*/
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
memset(hash, 0, sizeof(hash));
Convert to if_foreach_llmaddr() KPI.
2019-10-21 18:06:57 +00:00
if_foreach_llmaddr(ifp, gem_hash_maddr, hash);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
v |= GEM_MAC_RX_HASH_FILTER;
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
/* Now load the hash table into the chip (if we are using it). */
for (i = 0; i < 16; i++)
Split the registers into two halves in preparation for SBus support. Obtained from: NetBSD (loosely) MFC after: 2 weeks
2008-03-24 17:23:53 +00:00
GEM_BANK1_WRITE_4(sc,
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
GEM_MAC_HASH0 + i * (GEM_MAC_HASH1 - GEM_MAC_HASH0),
Incorporate changes made to the NetBSD version of this driver. - Remove some obsolete code (NetBSD gem.c r1.12) - Clean up how the local MAC address is programmed (NetBSD gem.c r1.13) - Make the driver work on PowerMacs with gigabit interfaces (NetBSD gem.c r1.14 and r1.15, gemreg.h r1.3 and r1.4, gemvar.h r1.6 and 1.7) - Suppress RX_MAC interrutps regarding the FRAME_COUNT register. (NetBSD gem.c r1.16 and r1.17) - Fix receiver lockups. (NetBSD gem.c r1.18, gemvar.h r1.8) - Distinguish between Apple and Sun variants (NetBSD if_gem_pci.c r1.9) Reviewed by: tmm Obtained from: NetBSD
2002-07-10 10:24:23 +00:00
hash[i]);
o In preparation for basing a new driver on this one: - ANSIfy - try to bring closer to style(9) - remove banal comments. o Add my copyright for having done lots of fixes and improvements.
2007-12-30 01:32:03 +00:00
chipit:
- In gem_reset_rx() also reset the RX MAC which is necessary in order to get it out of a stuck condition that can be caused by GEM_MAC_RX_OVERFLOW. - In gem_reset_rxdma() call gem_setladrf() in order to reprogram the RX filter and restore the previous content of GEM_MAC_RX_CONFIG. While at it consistently use the newly introduced sc_mac_rxcfg throughout the driver instead of reading the its old content. - Increment if_iqdrops instead of if_ierrors in case of RX buffer allocation failure. - According to the GEM datasheet the RX MAC should also be disabled in gem_setladrf() before changing its configuration. - Add error messages to gem_disable_{r,t}x() and take advantage of these throughout the driver instead of duplicating their functionality all over the place. In joint forces with: yongari
2011-06-28 16:16:43 +00:00
sc->sc_mac_rxcfg = v;
Correct r223648; as gem_init_locked() was calling gem_setladrf(), which sets GEM_MAC_RX_CONFIG based on sc_mac_rxcfg which in turn is initialized to zero, before reading the supposedly default configuration we were effectively not basing sc_mac_rxcfg and thus GEM_MAC_RX_CONFIG on the default configuration. Solve this by calling gem_setladrf() after reading in the default configuration of GEM_MAC_RX_CONFIG. This also avoids the need to distinguish whether gem_setladrf() should enable the RX MAC again and should be slightly more correct as we're now doing all of the RX MAC configuration in the intended step.
2011-07-12 13:06:44 +00:00
GEM_BANK1_WRITE_4(sc, GEM_MAC_RX_CONFIG, v | GEM_MAC_RX_ENABLE);
Add a driver for the Sun GEM (Gigabit) and ERI (100 Mb/s) PCI ethernet adaptors, ported from NetBSD.
2002-02-27 17:41:06 +00:00
}
Reference in New Issue Copy Permalink