2005-01-06 01:43:34 +00:00
|
|
|
/*-
|
2002-02-27 17:41:06 +00:00
|
|
|
* Copyright (C) 2001 Eduardo Horvath.
|
|
|
|
* 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.
|
|
|
|
*
|
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
|
|
|
* from: NetBSD: gemreg.h,v 1.8 2005/12/11 12:21:26 christos Exp
|
2002-02-27 17:41:06 +00:00
|
|
|
*
|
|
|
|
* $FreeBSD$
|
|
|
|
*/
|
|
|
|
|
|
|
|
#ifndef _IF_GEMREG_H
|
|
|
|
#define _IF_GEMREG_H
|
|
|
|
|
|
|
|
/* Register definitions for Sun GEM gigabit ethernet */
|
|
|
|
|
|
|
|
#define GEM_SEB_STATE 0x0000 /* SEB state reg, R/O */
|
|
|
|
#define GEM_CONFIG 0x0004 /* config reg */
|
|
|
|
#define GEM_STATUS 0x000c /* status reg */
|
2007-12-30 01:32:03 +00:00
|
|
|
/* Note: Reading the status reg clears bits 0-6. */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_INTMASK 0x0010
|
|
|
|
#define GEM_INTACK 0x0014 /* Interrupt acknowledge, W/O */
|
|
|
|
#define GEM_STATUS_ALIAS 0x001c
|
|
|
|
/* This is the same as the GEM_STATUS reg but reading it does not clear bits. */
|
|
|
|
#define GEM_ERROR_STATUS 0x1000 /* PCI error status R/C */
|
|
|
|
#define GEM_ERROR_MASK 0x1004
|
|
|
|
#define GEM_BIF_CONFIG 0x1008 /* BIF config reg */
|
|
|
|
#define GEM_BIF_DIAG 0x100c
|
|
|
|
#define GEM_RESET 0x1010 /* Software reset register */
|
|
|
|
|
|
|
|
|
|
|
|
/* Bits in GEM_SEB register */
|
|
|
|
#define GEM_SEB_ARB 0x000000002 /* Arbitration status */
|
|
|
|
#define GEM_SEB_RXWON 0x000000004
|
|
|
|
|
|
|
|
|
|
|
|
/* Bits in GEM_CONFIG register */
|
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
|
|
|
#define GEM_CONFIG_BURST_64 0x000000000 /* maximum burst size 64KB */
|
|
|
|
#define GEM_CONFIG_BURST_INF 0x000000001 /* infinite for entire packet */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_CONFIG_TXDMA_LIMIT 0x00000003e
|
|
|
|
#define GEM_CONFIG_RXDMA_LIMIT 0x0000007c0
|
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_CONFIG_RONPAULBIT and GEM_CONFIG_BUG2FIX are Apple only. */
|
|
|
|
#define GEM_CONFIG_RONPAULBIT 0x000000800 /* after infinite burst use */
|
|
|
|
/* memory read multiple for */
|
|
|
|
/* PCI commands */
|
|
|
|
#define GEM_CONFIG_BUG2FIX 0x000001000 /* fix RX hang after overflow */
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
#define GEM_CONFIG_TXDMA_LIMIT_SHIFT 1
|
|
|
|
#define GEM_CONFIG_RXDMA_LIMIT_SHIFT 6
|
|
|
|
|
|
|
|
|
|
|
|
/* Top part of GEM_STATUS has TX completion information */
|
|
|
|
#define GEM_STATUS_TX_COMPL 0xfff800000 /* TX completion 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
|
|
|
/*
|
2007-12-30 01:32:03 +00:00
|
|
|
* Interrupt bits, for both the GEM_STATUS and GEM_INTMASK regs
|
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
|
|
|
* Bits 0-6 auto-clear when read.
|
|
|
|
*/
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_INTR_TX_INTME 0x000000001 /* Frame w/INTME bit set sent */
|
|
|
|
#define GEM_INTR_TX_EMPTY 0x000000002 /* TX ring empty */
|
|
|
|
#define GEM_INTR_TX_DONE 0x000000004 /* TX complete */
|
|
|
|
#define GEM_INTR_RX_DONE 0x000000010 /* Got a packet */
|
|
|
|
#define GEM_INTR_RX_NOBUF 0x000000020
|
|
|
|
#define GEM_INTR_RX_TAG_ERR 0x000000040
|
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
|
|
|
#define GEM_INTR_PERR 0x000000080 /* Parity error */
|
|
|
|
#define GEM_INTR_PCS 0x000002000 /* Physical Code Sub-layer */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_INTR_TX_MAC 0x000004000
|
|
|
|
#define GEM_INTR_RX_MAC 0x000008000
|
|
|
|
#define GEM_INTR_MAC_CONTROL 0x000010000 /* MAC control interrupt */
|
|
|
|
#define GEM_INTR_MIF 0x000020000
|
|
|
|
#define GEM_INTR_BERR 0x000040000 /* Bus error interrupt */
|
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
|
|
|
#define GEM_INTR_BITS "\177\020" \
|
2002-02-27 17:41:06 +00:00
|
|
|
"b\0INTME\0b\1TXEMPTY\0b\2TXDONE\0" \
|
|
|
|
"b\4RXDONE\0b\5RXNOBUF\0b\6RX_TAG_ERR\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
|
|
|
"b\xdPCS\0b\xeTXMAC\0b\xfRXMAC\0" \
|
|
|
|
"b\x10MAC_CONTROL\0b\x11MIF\0b\x12IBERR\0\0"
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_ERROR_STATUS and GEM_ERROR_MASK PCI error bits */
|
|
|
|
#define GEM_ERROR_STAT_BADACK 0x000000001 /* No ACK64# */
|
|
|
|
#define GEM_ERROR_STAT_DTRTO 0x000000002 /* Delayed xaction timeout */
|
|
|
|
#define GEM_ERROR_STAT_OTHERS 0x000000004
|
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
|
|
|
#define GEM_ERROR_BITS "\177\020b\0ACKBAD\0b\1DTRTO\0b\2OTHER\0\0"
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_BIF_CONFIG register bits */
|
|
|
|
#define GEM_BIF_CONFIG_SLOWCLK 0x000000001 /* Parity error timing */
|
|
|
|
#define GEM_BIF_CONFIG_HOST_64 0x000000002 /* 64-bit host */
|
|
|
|
#define GEM_BIF_CONFIG_B64D_DIS 0x000000004 /* no 64-bit data cycle */
|
|
|
|
#define GEM_BIF_CONFIG_M66EN 0x000000008
|
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
|
|
|
#define GEM_BIF_CONFIG_BITS "\177\020b\0SLOWCLK\0b\1HOST64\0" \
|
|
|
|
"b\2B64DIS\0b\3M66EN\0\0"
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_RESET register bits -- TX and RX self clear when complete. */
|
|
|
|
#define GEM_RESET_TX 0x000000001 /* Reset TX half */
|
|
|
|
#define GEM_RESET_RX 0x000000002 /* Reset RX half */
|
|
|
|
#define GEM_RESET_RSTOUT 0x000000004 /* Force PCI RSTOUT# */
|
|
|
|
|
|
|
|
|
|
|
|
/* GEM TX DMA registers */
|
|
|
|
#define GEM_TX_KICK 0x2000 /* Write last valid desc + 1 */
|
|
|
|
#define GEM_TX_CONFIG 0x2004
|
|
|
|
#define GEM_TX_RING_PTR_LO 0x2008
|
|
|
|
#define GEM_TX_RING_PTR_HI 0x200c
|
|
|
|
|
|
|
|
#define GEM_TX_FIFO_WR_PTR 0x2014 /* FIFO write pointer */
|
|
|
|
#define GEM_TX_FIFO_SDWR_PTR 0x2018 /* FIFO shadow write pointer */
|
|
|
|
#define GEM_TX_FIFO_RD_PTR 0x201c /* FIFO read pointer */
|
|
|
|
#define GEM_TX_FIFO_SDRD_PTR 0x2020 /* FIFO shadow read pointer */
|
|
|
|
#define GEM_TX_FIFO_PKT_CNT 0x2024 /* FIFO packet counter */
|
|
|
|
|
|
|
|
#define GEM_TX_STATE_MACHINE 0x2028 /* ETX state machine reg */
|
|
|
|
#define GEM_TX_DATA_PTR_LO 0x2030
|
|
|
|
#define GEM_TX_DATA_PTR_HI 0x2034
|
|
|
|
|
|
|
|
#define GEM_TX_COMPLETION 0x2100
|
|
|
|
#define GEM_TX_FIFO_ADDRESS 0x2104
|
|
|
|
#define GEM_TX_FIFO_TAG 0x2108
|
|
|
|
#define GEM_TX_FIFO_DATA_LO 0x210c
|
|
|
|
#define GEM_TX_FIFO_DATA_HI_T1 0x2110
|
|
|
|
#define GEM_TX_FIFO_DATA_HI_T0 0x2114
|
|
|
|
#define GEM_TX_FIFO_SIZE 0x2118
|
|
|
|
#define GEM_TX_DEBUG 0x3028
|
|
|
|
|
|
|
|
|
2007-12-30 01:32:03 +00:00
|
|
|
/* GEM_TX_CONFIG register bits */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_TX_CONFIG_TXDMA_EN 0x00000001 /* TX DMA enable */
|
|
|
|
#define GEM_TX_CONFIG_TXRING_SZ 0x0000001e /* TX ring size */
|
|
|
|
#define GEM_TX_CONFIG_TXFIFO_TH 0x001ffc00 /* TX fifo threshold */
|
|
|
|
#define GEM_TX_CONFIG_PACED 0x00200000 /* TX_all_int modifier */
|
|
|
|
|
|
|
|
#define GEM_RING_SZ_32 (0<<1) /* 32 descriptors */
|
|
|
|
#define GEM_RING_SZ_64 (1<<1)
|
|
|
|
#define GEM_RING_SZ_128 (2<<1)
|
|
|
|
#define GEM_RING_SZ_256 (3<<1)
|
|
|
|
#define GEM_RING_SZ_512 (4<<1)
|
|
|
|
#define GEM_RING_SZ_1024 (5<<1)
|
|
|
|
#define GEM_RING_SZ_2048 (6<<1)
|
|
|
|
#define GEM_RING_SZ_4096 (7<<1)
|
|
|
|
#define GEM_RING_SZ_8192 (8<<1)
|
|
|
|
|
|
|
|
|
|
|
|
/* GEM_TX_COMPLETION register bits */
|
|
|
|
#define GEM_TX_COMPLETION_MASK 0x00001fff /* # of last descriptor */
|
|
|
|
|
|
|
|
|
|
|
|
/* GEM RX DMA registers */
|
|
|
|
#define GEM_RX_CONFIG 0x4000
|
|
|
|
#define GEM_RX_RING_PTR_LO 0x4004 /* 64-bits unaligned GAK! */
|
|
|
|
#define GEM_RX_RING_PTR_HI 0x4008 /* 64-bits unaligned GAK! */
|
|
|
|
|
|
|
|
#define GEM_RX_FIFO_WR_PTR 0x400c /* FIFO write pointer */
|
|
|
|
#define GEM_RX_FIFO_SDWR_PTR 0x4010 /* FIFO shadow write pointer */
|
|
|
|
#define GEM_RX_FIFO_RD_PTR 0x4014 /* FIFO read pointer */
|
|
|
|
#define GEM_RX_FIFO_PKT_CNT 0x4018 /* FIFO packet counter */
|
|
|
|
|
|
|
|
#define GEM_RX_STATE_MACHINE 0x401c /* ERX state machine reg */
|
|
|
|
#define GEM_RX_PAUSE_THRESH 0x4020
|
|
|
|
|
|
|
|
#define GEM_RX_DATA_PTR_LO 0x4024 /* ERX state machine reg */
|
|
|
|
#define GEM_RX_DATA_PTR_HI 0x4028 /* Damn thing is unaligned */
|
|
|
|
|
|
|
|
#define GEM_RX_KICK 0x4100 /* Write last valid desc + 1 */
|
|
|
|
#define GEM_RX_COMPLETION 0x4104 /* First pending desc */
|
|
|
|
#define GEM_RX_BLANKING 0x4108 /* Interrupt blanking reg */
|
|
|
|
|
|
|
|
#define GEM_RX_FIFO_ADDRESS 0x410c
|
|
|
|
#define GEM_RX_FIFO_TAG 0x4110
|
|
|
|
#define GEM_RX_FIFO_DATA_LO 0x4114
|
|
|
|
#define GEM_RX_FIFO_DATA_HI_T1 0x4118
|
|
|
|
#define GEM_RX_FIFO_DATA_HI_T0 0x411c
|
|
|
|
#define GEM_RX_FIFO_SIZE 0x4120
|
|
|
|
|
|
|
|
|
2007-12-30 01:32:03 +00:00
|
|
|
/* GEM_RX_CONFIG register bits */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_RX_CONFIG_RXDMA_EN 0x00000001 /* RX DMA enable */
|
|
|
|
#define GEM_RX_CONFIG_RXRING_SZ 0x0000001e /* RX ring size */
|
|
|
|
#define GEM_RX_CONFIG_BATCH_DIS 0x00000020 /* desc batching disable */
|
|
|
|
#define GEM_RX_CONFIG_FBOFF 0x00001c00 /* first byte offset */
|
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
|
|
|
#define GEM_RX_CONFIG_CXM_START 0x000fe000 /* cksum start offset bytes */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_RX_CONFIG_FIFO_THRS 0x07000000 /* fifo threshold size */
|
|
|
|
|
|
|
|
#define GEM_THRSH_64 0
|
|
|
|
#define GEM_THRSH_128 1
|
|
|
|
#define GEM_THRSH_256 2
|
|
|
|
#define GEM_THRSH_512 3
|
|
|
|
#define GEM_THRSH_1024 4
|
|
|
|
#define GEM_THRSH_2048 5
|
|
|
|
|
|
|
|
#define GEM_RX_CONFIG_FIFO_THRS_SHIFT 24
|
|
|
|
#define GEM_RX_CONFIG_FBOFF_SHFT 10
|
|
|
|
#define GEM_RX_CONFIG_CXM_START_SHFT 13
|
|
|
|
|
|
|
|
|
|
|
|
/* GEM_RX_PAUSE_THRESH register bits -- sizes in multiples of 64 bytes */
|
|
|
|
#define GEM_RX_PTH_XOFF_THRESH 0x000001ff
|
2002-07-10 10:24:23 +00:00
|
|
|
#define GEM_RX_PTH_XON_THRESH 0x001ff000
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_RX_BLANKING register bits */
|
|
|
|
#define GEM_RX_BLANKING_PACKETS 0x000001ff /* Delay intr for x packets */
|
2002-07-10 10:24:23 +00:00
|
|
|
#define GEM_RX_BLANKING_TIME 0x000ff000 /* Delay intr for x ticks */
|
|
|
|
#define GEM_RX_BLANKING_TIME_SHIFT 12
|
|
|
|
/* One tick is 2048 PCI clocks, or 16us at 66MHz */
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MAC registers */
|
|
|
|
#define GEM_MAC_TXRESET 0x6000 /* Store 1, cleared when done */
|
|
|
|
#define GEM_MAC_RXRESET 0x6004 /* ditto */
|
|
|
|
#define GEM_MAC_SEND_PAUSE_CMD 0x6008
|
|
|
|
#define GEM_MAC_TX_STATUS 0x6010
|
|
|
|
#define GEM_MAC_RX_STATUS 0x6014
|
|
|
|
#define GEM_MAC_CONTROL_STATUS 0x6018 /* MAC control status reg */
|
|
|
|
#define GEM_MAC_TX_MASK 0x6020 /* TX MAC mask register */
|
|
|
|
#define GEM_MAC_RX_MASK 0x6024
|
|
|
|
#define GEM_MAC_CONTROL_MASK 0x6028
|
|
|
|
#define GEM_MAC_TX_CONFIG 0x6030
|
|
|
|
#define GEM_MAC_RX_CONFIG 0x6034
|
|
|
|
#define GEM_MAC_CONTROL_CONFIG 0x6038
|
|
|
|
#define GEM_MAC_XIF_CONFIG 0x603c
|
|
|
|
#define GEM_MAC_IPG0 0x6040 /* inter packet gap 0 */
|
|
|
|
#define GEM_MAC_IPG1 0x6044 /* inter packet gap 1 */
|
|
|
|
#define GEM_MAC_IPG2 0x6048 /* inter packet gap 2 */
|
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
|
|
|
#define GEM_MAC_SLOT_TIME 0x604c /* slot time, bits 0-7 */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MAC_MAC_MIN_FRAME 0x6050
|
|
|
|
#define GEM_MAC_MAC_MAX_FRAME 0x6054
|
|
|
|
#define GEM_MAC_PREAMBLE_LEN 0x6058
|
|
|
|
#define GEM_MAC_JAM_SIZE 0x605c
|
|
|
|
#define GEM_MAC_ATTEMPT_LIMIT 0x6060
|
|
|
|
#define GEM_MAC_CONTROL_TYPE 0x6064
|
|
|
|
|
|
|
|
#define GEM_MAC_ADDR0 0x6080 /* Normal MAC address 0 */
|
|
|
|
#define GEM_MAC_ADDR1 0x6084
|
|
|
|
#define GEM_MAC_ADDR2 0x6088
|
|
|
|
#define GEM_MAC_ADDR3 0x608c /* Alternate MAC address 0 */
|
|
|
|
#define GEM_MAC_ADDR4 0x6090
|
|
|
|
#define GEM_MAC_ADDR5 0x6094
|
|
|
|
#define GEM_MAC_ADDR6 0x6098 /* Control MAC address 0 */
|
|
|
|
#define GEM_MAC_ADDR7 0x609c
|
|
|
|
#define GEM_MAC_ADDR8 0x60a0
|
|
|
|
|
|
|
|
#define GEM_MAC_ADDR_FILTER0 0x60a4
|
|
|
|
#define GEM_MAC_ADDR_FILTER1 0x60a8
|
|
|
|
#define GEM_MAC_ADDR_FILTER2 0x60ac
|
|
|
|
#define GEM_MAC_ADR_FLT_MASK1_2 0x60b0 /* Address filter mask 1,2 */
|
|
|
|
#define GEM_MAC_ADR_FLT_MASK0 0x60b4 /* Address filter mask 0 reg */
|
|
|
|
|
|
|
|
#define GEM_MAC_HASH0 0x60c0 /* Hash table 0 */
|
|
|
|
#define GEM_MAC_HASH1 0x60c4
|
|
|
|
#define GEM_MAC_HASH2 0x60c8
|
|
|
|
#define GEM_MAC_HASH3 0x60cc
|
|
|
|
#define GEM_MAC_HASH4 0x60d0
|
|
|
|
#define GEM_MAC_HASH5 0x60d4
|
|
|
|
#define GEM_MAC_HASH6 0x60d8
|
|
|
|
#define GEM_MAC_HASH7 0x60dc
|
|
|
|
#define GEM_MAC_HASH8 0x60e0
|
|
|
|
#define GEM_MAC_HASH9 0x60e4
|
|
|
|
#define GEM_MAC_HASH10 0x60e8
|
|
|
|
#define GEM_MAC_HASH11 0x60ec
|
|
|
|
#define GEM_MAC_HASH12 0x60f0
|
|
|
|
#define GEM_MAC_HASH13 0x60f4
|
|
|
|
#define GEM_MAC_HASH14 0x60f8
|
|
|
|
#define GEM_MAC_HASH15 0x60fc
|
|
|
|
|
|
|
|
#define GEM_MAC_NORM_COLL_CNT 0x6100 /* Normal collision counter */
|
|
|
|
#define GEM_MAC_FIRST_COLL_CNT 0x6104 /* 1st successful collision cntr */
|
|
|
|
#define GEM_MAC_EXCESS_COLL_CNT 0x6108 /* Excess collision counter */
|
|
|
|
#define GEM_MAC_LATE_COLL_CNT 0x610c /* Late collision counter */
|
|
|
|
#define GEM_MAC_DEFER_TMR_CNT 0x6110 /* defer timer counter */
|
|
|
|
#define GEM_MAC_PEAK_ATTEMPTS 0x6114
|
|
|
|
#define GEM_MAC_RX_FRAME_COUNT 0x6118
|
|
|
|
#define GEM_MAC_RX_LEN_ERR_CNT 0x611c
|
|
|
|
#define GEM_MAC_RX_ALIGN_ERR 0x6120
|
|
|
|
#define GEM_MAC_RX_CRC_ERR_CNT 0x6124
|
|
|
|
#define GEM_MAC_RX_CODE_VIOL 0x6128
|
|
|
|
#define GEM_MAC_RANDOM_SEED 0x6130
|
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
|
|
|
#define GEM_MAC_MAC_STATE 0x6134 /* MAC state machine reg */
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MAC_SEND_PAUSE_CMD register bits */
|
|
|
|
#define GEM_MAC_PAUSE_CMD_TIME 0x0000ffff
|
|
|
|
#define GEM_MAC_PAUSE_CMD_SEND 0x00010000
|
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MAC_TX_STATUS and _MASK register bits */
|
|
|
|
#define GEM_MAC_TX_XMIT_DONE 0x00000001
|
|
|
|
#define GEM_MAC_TX_UNDERRUN 0x00000002
|
|
|
|
#define GEM_MAC_TX_PKT_TOO_LONG 0x00000004
|
|
|
|
#define GEM_MAC_TX_NCC_EXP 0x00000008 /* Normal collision cnt exp */
|
|
|
|
#define GEM_MAC_TX_ECC_EXP 0x00000010
|
|
|
|
#define GEM_MAC_TX_LCC_EXP 0x00000020
|
|
|
|
#define GEM_MAC_TX_FCC_EXP 0x00000040
|
|
|
|
#define GEM_MAC_TX_DEFER_EXP 0x00000080
|
|
|
|
#define GEM_MAC_TX_PEAK_EXP 0x00000100
|
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MAC_RX_STATUS and _MASK register bits */
|
|
|
|
#define GEM_MAC_RX_DONE 0x00000001
|
|
|
|
#define GEM_MAC_RX_OVERFLOW 0x00000002
|
|
|
|
#define GEM_MAC_RX_FRAME_CNT 0x00000004
|
|
|
|
#define GEM_MAC_RX_ALIGN_EXP 0x00000008
|
|
|
|
#define GEM_MAC_RX_CRC_EXP 0x00000010
|
|
|
|
#define GEM_MAC_RX_LEN_EXP 0x00000020
|
|
|
|
#define GEM_MAC_RX_CVI_EXP 0x00000040 /* Code violation */
|
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MAC_CONTROL_STATUS and GEM_MAC_CONTROL_MASK register bits */
|
|
|
|
#define GEM_MAC_PAUSED 0x00000001 /* Pause received */
|
|
|
|
#define GEM_MAC_PAUSE 0x00000002 /* enter pause state */
|
|
|
|
#define GEM_MAC_RESUME 0x00000004 /* exit pause state */
|
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
|
|
|
#define GEM_MAC_PAUSE_TIME_SLTS 0xffff0000 /* pause time in slots */
|
|
|
|
#define GEM_MAC_STATUS_BITS "\177\020b\0PAUSED\0b\1PAUSE\0b\2RESUME\0\0"
|
|
|
|
|
|
|
|
#define GEM_MAC_PAUSE_TIME_SHFT 16
|
|
|
|
#define GEM_MAC_PAUSE_TIME(x) \
|
|
|
|
(((x) & GEM_MAC_PAUSE_TIME_SLTS) >> GEM_MAC_PAUSE_TIME_SHFT)
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
/* GEM_MAC_XIF_CONFIG register bits */
|
|
|
|
#define GEM_MAC_XIF_TX_MII_ENA 0x00000001 /* Enable XIF output drivers */
|
|
|
|
#define GEM_MAC_XIF_MII_LOOPBK 0x00000002 /* Enable MII loopback mode */
|
|
|
|
#define GEM_MAC_XIF_ECHO_DISABL 0x00000004 /* Disable echo */
|
2002-07-10 10:24:23 +00:00
|
|
|
#define GEM_MAC_XIF_GMII_MODE 0x00000008 /* Select GMII/MII mode */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MAC_XIF_MII_BUF_ENA 0x00000010 /* Enable MII recv buffers */
|
|
|
|
#define GEM_MAC_XIF_LINK_LED 0x00000020 /* force link LED active */
|
|
|
|
#define GEM_MAC_XIF_FDPLX_LED 0x00000040 /* force FDPLX LED active */
|
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
|
|
|
#define GEM_MAC_XIF_BITS "\177\020b\0TXMIIENA\0b\1MIILOOP\0b\2NOECHO" \
|
|
|
|
"\0b\3GMII\0b\4MIIBUFENA\0b\5LINKLED\0" \
|
|
|
|
"b\6FDLED\0\0"
|
|
|
|
|
|
|
|
/*
|
|
|
|
* GEM_MAC_SLOT_TIME register
|
|
|
|
* The slot time is used as PAUSE time unit, value depends on whether carrier
|
|
|
|
* extension is enabled.
|
|
|
|
*/
|
|
|
|
#define GEM_MAC_SLOT_TIME_CARR_EXTEND 0x200
|
|
|
|
#define GEM_MAC_SLOT_TIME_NORMAL 0x40
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
/* GEM_MAC_TX_CONFIG register bits */
|
|
|
|
#define GEM_MAC_TX_ENABLE 0x00000001 /* TX enable */
|
|
|
|
#define GEM_MAC_TX_IGN_CARRIER 0x00000002 /* Ignore carrier sense */
|
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
|
|
|
#define GEM_MAC_TX_IGN_COLLIS 0x00000004 /* ignore collisions */
|
2007-12-30 01:32:03 +00:00
|
|
|
#define GEM_MAC_TX_ENA_IPG0 0x00000008 /* extend RX-to-TX IPG */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MAC_TX_NGU 0x00000010 /* Never give up */
|
|
|
|
#define GEM_MAC_TX_NGU_LIMIT 0x00000020 /* Never give up limit */
|
|
|
|
#define GEM_MAC_TX_NO_BACKOFF 0x00000040
|
|
|
|
#define GEM_MAC_TX_SLOWDOWN 0x00000080
|
|
|
|
#define GEM_MAC_TX_NO_FCS 0x00000100 /* no FCS will be generated */
|
|
|
|
#define GEM_MAC_TX_CARR_EXTEND 0x00000200 /* Ena TX Carrier Extension */
|
|
|
|
/* Carrier Extension is required for half duplex Gbps operation */
|
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
|
|
|
#define GEM_MAC_TX_CONFIG_BITS "\177\020" \
|
|
|
|
"b\0TXENA\0b\1IGNCAR\0b\2IGNCOLLIS\0" \
|
|
|
|
"b\3IPG0ENA\0b\4TXNGU\0b\5TXNGULIM\0" \
|
|
|
|
"b\6NOBKOFF\0b\7SLOWDN\0b\x8NOFCS\0" \
|
|
|
|
"b\x9TXCARREXT\0\0"
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MAC_RX_CONFIG register bits */
|
|
|
|
#define GEM_MAC_RX_ENABLE 0x00000001 /* RX enable */
|
|
|
|
#define GEM_MAC_RX_STRIP_PAD 0x00000002 /* strip pad bytes */
|
|
|
|
#define GEM_MAC_RX_STRIP_CRC 0x00000004
|
|
|
|
#define GEM_MAC_RX_PROMISCUOUS 0x00000008 /* promiscuous mode */
|
|
|
|
#define GEM_MAC_RX_PROMISC_GRP 0x00000010 /* promiscuous group mode */
|
|
|
|
#define GEM_MAC_RX_HASH_FILTER 0x00000020 /* enable hash filter */
|
|
|
|
#define GEM_MAC_RX_ADDR_FILTER 0x00000040 /* enable address filter */
|
|
|
|
#define GEM_MAC_RX_ERRCHK_DIS 0x00000080 /* disable error checking */
|
|
|
|
#define GEM_MAC_RX_CARR_EXTEND 0x00000100 /* Ena RX Carrier Extension */
|
|
|
|
/*
|
|
|
|
* Carrier Extension enables reception of packet bursts generated by
|
|
|
|
* senders with carrier extension enabled.
|
|
|
|
*/
|
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
|
|
|
#define GEM_MAC_RX_CONFIG_BITS "\177\020" \
|
|
|
|
"b\0RXENA\0b\1STRPAD\0b\2STRCRC\0" \
|
|
|
|
"b\3PROMIS\0b\4PROMISCGRP\0b\5HASHFLTR\0" \
|
|
|
|
"b\6ADDRFLTR\0b\7ERRCHKDIS\0b\x9TXCARREXT\0\0"
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MAC_CONTROL_CONFIG bits */
|
|
|
|
#define GEM_MAC_CC_TX_PAUSE 0x00000001 /* send pause enabled */
|
|
|
|
#define GEM_MAC_CC_RX_PAUSE 0x00000002 /* receive pause enabled */
|
|
|
|
#define GEM_MAC_CC_PASS_PAUSE 0x00000004 /* pass pause up */
|
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
|
|
|
#define GEM_MAC_CC_BITS "\177\020b\0TXPAUSE\0b\1RXPAUSE\0b\2NOPAUSE\0\0"
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM MIF registers */
|
2007-12-30 01:32:03 +00:00
|
|
|
/* Bit bang registers use low bit only. */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MIF_BB_CLOCK 0x6200 /* bit bang clock */
|
|
|
|
#define GEM_MIF_BB_DATA 0x6204 /* bit bang data */
|
|
|
|
#define GEM_MIF_BB_OUTPUT_ENAB 0x6208
|
|
|
|
#define GEM_MIF_FRAME 0x620c /* MIF frame - ctl and data */
|
|
|
|
#define GEM_MIF_CONFIG 0x6210
|
|
|
|
#define GEM_MIF_INTERRUPT_MASK 0x6214
|
|
|
|
#define GEM_MIF_BASIC_STATUS 0x6218
|
|
|
|
#define GEM_MIF_STATE_MACHINE 0x621c
|
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MIF_FRAME bits */
|
|
|
|
#define GEM_MIF_FRAME_DATA 0x0000ffff
|
|
|
|
#define GEM_MIF_FRAME_TA0 0x00010000 /* TA bit, 1 for completion */
|
|
|
|
#define GEM_MIF_FRAME_TA1 0x00020000 /* TA bits */
|
|
|
|
#define GEM_MIF_FRAME_REG_ADDR 0x007c0000
|
|
|
|
#define GEM_MIF_FRAME_PHY_ADDR 0x0f800000 /* phy address, should be 0 */
|
|
|
|
#define GEM_MIF_FRAME_OP 0x30000000 /* operation - write/read */
|
|
|
|
#define GEM_MIF_FRAME_START 0xc0000000 /* START bits */
|
|
|
|
|
|
|
|
#define GEM_MIF_FRAME_READ 0x60020000
|
|
|
|
#define GEM_MIF_FRAME_WRITE 0x50020000
|
|
|
|
|
|
|
|
#define GEM_MIF_REG_SHIFT 18
|
|
|
|
#define GEM_MIF_PHY_SHIFT 23
|
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MIF_CONFIG register bits */
|
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
|
|
|
#define GEM_MIF_CONFIG_PHY_SEL 0x00000001 /* PHY select, 0=MDIO0 */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MIF_CONFIG_POLL_ENA 0x00000002 /* poll enable */
|
|
|
|
#define GEM_MIF_CONFIG_BB_ENA 0x00000004 /* bit bang enable */
|
|
|
|
#define GEM_MIF_CONFIG_REG_ADR 0x000000f8 /* poll register address */
|
|
|
|
#define GEM_MIF_CONFIG_MDI0 0x00000100 /* MDIO_0 Data/MDIO_0 atached */
|
|
|
|
#define GEM_MIF_CONFIG_MDI1 0x00000200 /* MDIO_1 Data/MDIO_1 atached */
|
|
|
|
#define GEM_MIF_CONFIG_PHY_ADR 0x00007c00 /* poll PHY address */
|
2007-12-30 01:32:03 +00:00
|
|
|
/* MDI0 is the onboard transceiver, MDI1 is external, PHYAD for both is 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
|
|
|
#define GEM_MIF_CONFIG_BITS "\177\020b\0PHYSEL\0b\1POLL\0b\2BBENA\0" \
|
|
|
|
"b\x8MDIO0\0b\x9MDIO1\0\0"
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MIF_BASIC_STATUS and GEM_MIF_INTERRUPT_MASK bits */
|
|
|
|
#define GEM_MIF_STATUS 0x0000ffff
|
|
|
|
#define GEM_MIF_BASIC 0xffff0000
|
|
|
|
/*
|
|
|
|
* The Basic part is the last value read in the POLL field of the config
|
|
|
|
* register.
|
|
|
|
*
|
|
|
|
* The status part indicates the bits that have changed.
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
2007-12-30 01:32:03 +00:00
|
|
|
/* GEM PCS/Serial link registers */
|
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
|
|
|
/* DO NOT TOUCH THESE REGISTERS ON ERI -- IT HARD HANGS. */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MII_CONTROL 0x9000
|
|
|
|
#define GEM_MII_STATUS 0x9004
|
|
|
|
#define GEM_MII_ANAR 0x9008 /* MII advertisement 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
|
|
|
#define GEM_MII_ANLPAR 0x900c /* Link Partner Ability Reg */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MII_CONFIG 0x9010
|
|
|
|
#define GEM_MII_STATE_MACHINE 0x9014
|
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
|
|
|
#define GEM_MII_INTERRUP_STATUS 0x9018 /* PCS interrupt state */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MII_DATAPATH_MODE 0x9050
|
|
|
|
#define GEM_MII_SLINK_CONTROL 0x9054 /* Serial link control */
|
|
|
|
#define GEM_MII_OUTPUT_SELECT 0x9058
|
|
|
|
#define GEM_MII_SLINK_STATUS 0x905c /* serial link 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_MII_CONTROL bits - PCS "BMCR" (Basic Mode Control Reg) */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MII_CONTROL_RESET 0x00008000
|
|
|
|
#define GEM_MII_CONTROL_LOOPBK 0x00004000 /* 10-bit i/f loopback */
|
|
|
|
#define GEM_MII_CONTROL_1000M 0x00002000 /* speed select, always 0 */
|
|
|
|
#define GEM_MII_CONTROL_AUTONEG 0x00001000 /* auto negotiation enabled */
|
|
|
|
#define GEM_MII_CONTROL_POWERDN 0x00000800
|
|
|
|
#define GEM_MII_CONTROL_ISOLATE 0x00000400 /* isolate phy from mii */
|
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
|
|
|
#define GEM_MII_CONTROL_RAN 0x00000200 /* restart auto negotiation */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MII_CONTROL_FDUPLEX 0x00000100 /* full duplex, always 0 */
|
|
|
|
#define GEM_MII_CONTROL_COL_TST 0x00000080 /* collision test */
|
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
|
|
|
#define GEM_MII_CONTROL_BITS "\177\020b\7COLTST\0b\x8_FD\0b\x9RAN\0" \
|
|
|
|
"b\xaISOLATE\0b\xbPWRDWN\0b\xc_ANEG\0" \
|
|
|
|
"b\xdGIGE\0b\xeLOOP\0b\xfRESET\0\0"
|
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
|
|
|
/* GEM_MII_STATUS reg - PCS "BMSR" (Basic Mode Status Reg) */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MII_STATUS_GB_FDX 0x00000400 /* can perform GBit FDX */
|
|
|
|
#define GEM_MII_STATUS_GB_HDX 0x00000200 /* can perform GBit HDX */
|
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
|
|
|
#define GEM_MII_STATUS_UNK 0x00000100
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MII_STATUS_ANEG_CPT 0x00000020 /* auto negotiate compete */
|
|
|
|
#define GEM_MII_STATUS_REM_FLT 0x00000010 /* remote fault detected */
|
|
|
|
#define GEM_MII_STATUS_ACFG 0x00000008 /* can auto negotiate */
|
|
|
|
#define GEM_MII_STATUS_LINK_STS 0x00000004 /* link status */
|
|
|
|
#define GEM_MII_STATUS_JABBER 0x00000002 /* jabber condition detected */
|
|
|
|
#define GEM_MII_STATUS_EXTCAP 0x00000001 /* extended register capability */
|
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
|
|
|
#define GEM_MII_STATUS_BITS "\177\020b\0EXTCAP\0b\1JABBER\0b\2LINKSTS\0" \
|
|
|
|
"b\3ACFG\0b\4REMFLT\0b\5ANEGCPT\0b\x9GBHDX\0" \
|
|
|
|
"b\xaGBFDX\0\0"
|
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
|
|
|
/* GEM_MII_ANAR and GEM_MII_ANLPAR reg bits */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MII_ANEG_NP 0x00008000 /* next page bit */
|
|
|
|
#define GEM_MII_ANEG_ACK 0x00004000 /* ack reception of */
|
|
|
|
/* Link Partner Capability */
|
|
|
|
#define GEM_MII_ANEG_RF 0x00003000 /* advertise remote fault cap */
|
|
|
|
#define GEM_MII_ANEG_ASYM_PAUSE 0x00000100 /* asymmetric pause */
|
|
|
|
#define GEM_MII_ANEG_SYM_PAUSE 0x00000080 /* symmetric pause */
|
|
|
|
#define GEM_MII_ANEG_HLF_DUPLX 0x00000040
|
|
|
|
#define GEM_MII_ANEG_FUL_DUPLX 0x00000020
|
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
|
|
|
#define GEM_MII_ANEG_BITS "\177\020b\5FDX\0b\6HDX\0b\7SYMPAUSE\0" \
|
|
|
|
"\b\x8_ASYMPAUSE\0\b\xdREMFLT\0\b\xeLPACK\0" \
|
|
|
|
"\b\xfNPBIT\0\0"
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MII_CONFIG 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
|
|
|
#define GEM_MII_CONFIG_TIMER 0x0000000e /* link monitor timer values */
|
|
|
|
#define GEM_MII_CONFIG_ANTO 0x00000020 /* 10ms ANEG timer override */
|
|
|
|
#define GEM_MII_CONFIG_JS 0x00000018 /* Jitter Study, 0 normal
|
|
|
|
* 1 high freq, 2 low freq */
|
|
|
|
#define GEM_MII_CONFIG_SDL 0x00000004 /* Signal Detect active low */
|
|
|
|
#define GEM_MII_CONFIG_SDO 0x00000002 /* Signal Detect Override */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MII_CONFIG_ENABLE 0x00000001 /* Enable PCS */
|
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
|
|
|
#define GEM_MII_CONFIG_BITS "\177\020b\0PCSENA\0\0"
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* GEM_MII_STATE_MACHINE
|
|
|
|
* XXX These are best guesses from observed behavior.
|
|
|
|
*/
|
|
|
|
#define GEM_MII_FSM_STOP 0x00000000 /* stopped */
|
|
|
|
#define GEM_MII_FSM_RUN 0x00000001 /* running */
|
|
|
|
#define GEM_MII_FSM_UNKWN 0x00000100 /* unknown */
|
|
|
|
#define GEM_MII_FSM_DONE 0x00000101 /* complete */
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* GEM_MII_INTERRUP_STATUS reg
|
|
|
|
* No mask register; mask with the global interrupt mask register.
|
|
|
|
*/
|
|
|
|
#define GEM_MII_INTERRUP_LINK 0x00000004 /* PCS link status change */
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MII_DATAPATH_MODE reg */
|
|
|
|
#define GEM_MII_DATAPATH_SERIAL 0x00000001 /* Serial link */
|
|
|
|
#define GEM_MII_DATAPATH_SERDES 0x00000002 /* Use PCS via 10bit interfac */
|
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
|
|
|
#define GEM_MII_DATAPATH_MII 0x00000004 /* Use {G}MII, not PCS */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MII_DATAPATH_MIIOUT 0x00000008 /* enable serial output on GMII */
|
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
|
|
|
#define GEM_MII_DATAPATH_BITS "\177\020" \
|
|
|
|
"b\0SERIAL\0b\1SERDES\0b\2MII\0b\3MIIOUT\0\0"
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MII_SLINK_CONTROL 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
|
|
|
#define GEM_MII_SLINK_LOOPBACK 0x00000001 /* enable loopback at sl, logic
|
|
|
|
* reversed for SERDES */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_MII_SLINK_EN_SYNC_D 0x00000002 /* enable sync detection */
|
|
|
|
#define GEM_MII_SLINK_LOCK_REF 0x00000004 /* lock reference clock */
|
|
|
|
#define GEM_MII_SLINK_EMPHASIS 0x00000008 /* enable emphasis */
|
|
|
|
#define GEM_MII_SLINK_SELFTEST 0x000001c0
|
|
|
|
#define GEM_MII_SLINK_POWER_OFF 0x00000200 /* Power down serial link */
|
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
|
|
|
#define GEM_MII_SLINK_CONTROL_BITS \
|
|
|
|
"\177\020b\0LOOP\0b\1ENASYNC\0b\2LOCKREF" \
|
|
|
|
"\0b\3EMPHASIS\0b\x9PWRDWN\0\0"
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* GEM_MII_SLINK_STATUS reg */
|
|
|
|
#define GEM_MII_SLINK_TEST 0x00000000 /* undergoing test */
|
|
|
|
#define GEM_MII_SLINK_LOCKED 0x00000001 /* waiting 500us lockrefn */
|
|
|
|
#define GEM_MII_SLINK_COMMA 0x00000002 /* waiting for comma detect */
|
|
|
|
#define GEM_MII_SLINK_SYNC 0x00000003 /* recv data synchronized */
|
|
|
|
|
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
|
|
|
/*
|
|
|
|
* PCI Expansion ROM runtime access
|
|
|
|
* Sun GEMs map a 1MB space for the PCI Expansion ROM as the second half
|
|
|
|
* of the first register bank, although they only support up to 64KB ROMs.
|
|
|
|
*/
|
|
|
|
#define GEM_PCI_ROM_OFFSET 0x100000
|
|
|
|
#define GEM_PCI_ROM_SIZE 0x10000
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
/* Wired GEM PHY addresses */
|
|
|
|
#define GEM_PHYAD_INTERNAL 1
|
|
|
|
#define GEM_PHYAD_EXTERNAL 0
|
|
|
|
|
|
|
|
/*
|
2007-12-30 01:32:03 +00:00
|
|
|
* GEM descriptor table structures
|
2002-02-27 17:41:06 +00:00
|
|
|
*/
|
|
|
|
struct gem_desc {
|
|
|
|
uint64_t gd_flags;
|
|
|
|
uint64_t gd_addr;
|
|
|
|
};
|
|
|
|
|
|
|
|
/* Transmit flags */
|
|
|
|
#define GEM_TD_BUFSIZE 0x0000000000007fffLL
|
|
|
|
#define GEM_TD_CXSUM_START 0x00000000001f8000LL /* Cxsum start offset */
|
2007-06-04 06:01:04 +00:00
|
|
|
#define GEM_TD_CXSUM_STARTSHFT 15
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_TD_CXSUM_STUFF 0x000000001fe00000LL /* Cxsum stuff offset */
|
2007-06-04 06:01:04 +00:00
|
|
|
#define GEM_TD_CXSUM_STUFFSHFT 21
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_TD_CXSUM_ENABLE 0x0000000020000000LL /* Cxsum generation enable */
|
|
|
|
#define GEM_TD_END_OF_PACKET 0x0000000040000000LL
|
|
|
|
#define GEM_TD_START_OF_PACKET 0x0000000080000000LL
|
|
|
|
#define GEM_TD_INTERRUPT_ME 0x0000000100000000LL /* Interrupt me now */
|
|
|
|
#define GEM_TD_NO_CRC 0x0000000200000000LL /* do not insert crc */
|
|
|
|
/*
|
|
|
|
* Only need to set GEM_TD_CXSUM_ENABLE, GEM_TD_CXSUM_STUFF,
|
|
|
|
* GEM_TD_CXSUM_START, and GEM_TD_INTERRUPT_ME in 1st descriptor of a group.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* Receive flags */
|
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
|
|
|
#define GEM_RD_CHECKSUM 0x000000000000ffffLL /* is the complement */
|
2002-02-27 17:41:06 +00:00
|
|
|
#define GEM_RD_BUFSIZE 0x000000007fff0000LL
|
|
|
|
#define GEM_RD_OWN 0x0000000080000000LL /* 1 - owned by h/w */
|
|
|
|
#define GEM_RD_HASHVAL 0x0ffff00000000000LL
|
|
|
|
#define GEM_RD_HASH_PASS 0x1000000000000000LL /* passed hash filter */
|
|
|
|
#define GEM_RD_ALTERNATE_MAC 0x2000000000000000LL /* Alternate MAC adrs */
|
|
|
|
#define GEM_RD_BAD_CRC 0x4000000000000000LL
|
|
|
|
|
|
|
|
#define GEM_RD_BUFSHIFT 16
|
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
|
|
|
#define GEM_RD_BUFLEN(x) (((x) & GEM_RD_BUFSIZE) >> GEM_RD_BUFSHIFT)
|
2002-02-27 17:41:06 +00:00
|
|
|
|
|
|
|
#endif
|